Decision making under risk in agriculture

Institut für Lebensmittel - und Ressourcenökonomik

Decision making under risk in

agriculture

–

An experimental and survey based analysis

Dissertation

zur Erlangung des Grades

Doktorin der Agrarwissenschaften (Dr. agr.)

der Landwirtschaftlichen Fakultät

der Rheinischen Friedrich-Wilhelms-Universität Bonn

von

Manuela Meraner

aus

Bozen, Italien

Bonn 2017

Referent: Professor Dr. Robert Finger

Korreferent: Professor Dr. Thomas Heckelei

Tag der mündlichen Prüfung: 08.09.2017

Danksagung

An dieser Stelle möchte ich allen danken, die zum Gelingen dieser Arbeit

beigetragen haben.

Besonderes möchte ich mich bei meinem Doktorvater Herrn Prof. Dr. Robert

Finger bedanken, der nicht nur die wissenschaftliche Betreuung meiner Arbeit

übernommen und deren Fortgang durch kritische und inspirierende Diskussionen

bereichert, sondern mir durch einen hohen Grad an Freiheit, in der inhaltlichen

Ausgestaltung und in meiner Arbeitsorganisation meine berufliche und persönliche

Weiterentwicklung stets gefördert hat.

Für die bereitwillige Übernahme des Zweitgutachtens bedanke ich mich herzlich

bei Herrn Prof. Dr. Thomas Heckelei. Durch die Anstellung am Lehrstuhl für

Agrarpolitik als studentische Hilfskraft 2012 konnte ich mich für das Double

Degree in Wageningen bewerben, womit der Grundstein für die nächsten fünf

Jahre gelegt wurde, vielen Dank!

Meinen Kolleginnen und Kollegen der AECP Gruppe an der ETH Zürich und vor

allem meinem Bürokollegen Niklas danke ich für die sowohl wissenschaftlichen

als auch privaten – Diskussionen, ermutigende Worte und gute Ratschläge. Des

Weiteren möchte ich mich bei meinen Freunden und Kolleginnen und Kollegen im

ILR bedanken die mich während meines Aufenthaltes in Bonn aber auch darüber

hinaus unterstützt haben.

Nicht zuletzt bedanke ich mich bei meiner Familie für die Verbundenheit mit der

Landwirtschaft, die sie mir mitgegeben haben und die Möglichkeiten und

Freiheiten meinen Interessen nachzugehen. Insbesondere bedanke ich mich bei Leo

der während meines Dissertationsprojektes Teil meiner Familie und wichtigste

moralische Stütze wurde.

„Okay, ladies, now let's get in formation, 'cause I slay”

Beyoncé Giselle Knowles-Carter

Abstract

Risk, and risk management are inherent to agricultural production. We distinguish

two categories of factors influencing the farmers’ decision on risk management

strategy: external and internal factors. Factors external to the farm holding include

the geographic, political and climatic environment. Internal factors include

personal characteristics and the farmers’ frame of reference. The presented studies

in this thesis include analysis of external and internal determinants of the farmers’

risk management decision in the German region North-Rhine-Westphalia focusing

on (i) effects of contextualizing experimental risk preference elicitation methods,

(ii) determinants of risk management choices amongst livestock farmers, (iii)

underlying motives of agritourism farmers and (iv) characteristics of diversifying

farmers in the peri-urban context.

Four unique case study analyses are conducted. First data collected from

agricultural students is used to explore effects of contextualization in experimental

risk preference elicitation methods. Second, data collected from livestock farmers

is used for a holistic analysis of determinants of risk management choices. Third,

agritourism and non-agritourism farmers are interviewed to explore underlying

motives of farmers engaging in agritourism activities, and to find distinguishing

farm, farmer and household characteristics of both groups. Fourth, data from

farmers in the Ruhr-metropolitan area are collected to analyze their diversification

decisions in the peri-urban context. The collected data is analyzed using

econometric methods.

The results show that contextualization of experimental risk preference elicitation

methods decreases misspecifications, and improves obtained data quality.

Furthermore, the analysis of farmers’ behavior under risk show that the individual

farmer’s frame of reference (i.e. risk preferences, perception and past experiences)

determines the choice of risk management strategy. Additionally, results show that

proximity to urban agglomerations incentivizes farmers to engage in on-farm, non-

agricultural diversification. Farmers’ decision to diversify is found to be

opportunity driven, leading to the conclusion that improved information with

regards to existing and new innovative diversification channels in the farming

context is essential to improve rural development.

Keywords: risk management, experimental risk preference elicitation methods,

risk perception, decision analysis, on-farm non-agricultural

diversification

Zusammenfassung

Das Risikomanagement ist wesentlicher Bestandteil der landwirtschaftlichen

Produktion. Es werden zwei Kategorien von Faktoren, die

Risikomanagemententscheidungen von Landwirten beeinflussen, unterschieden:

externe und interne Faktoren. Zu den externen Faktoren gehören geographische

Merkmale des Betriebs sowie Klimafaktoren oder das politische Umfeld. Interne

Faktoren sind persönliche Merkmale des Landwirts und der landwirtschaftlichen

Familie sowie der Entscheidungsrahmen der Landwirte. In den vorliegenden

Studien werden externe und interne Determinanten untersucht, welche die

Entscheidungsfindung über die Risikomanagementstrategien der Landwirte im

deutschen Bundesland Nordrhein-Westfalen beeinflussen, indem (i) Effekte der

Kontextualisierung von experimentellen Methoden zur Risikopräferenzmessung,

(ii) Determinanten von Risikomanagement-Entscheidungen bei

Veredelungsbetrieben, (iii) zugrundeliegende Motive von Landwirten mit

Agrartourismus sowie (iv) Charakteristika von diversifizierten Landwirten im peri-

urbanen Raum untersucht werden.

In der vorliegenden Arbeit werden vier einzigartige Fallstudien analysiert. Erstens,

werden Daten aus Experimenten mit landwirtschaftlichen Studenten verwendet, um

Effekte der Kontextualisierung in experimentellen Methoden zur

Risikopräferenzmessung zu erforschen. Zweitens werden Befragungsdaten von

Veredelungsbetrieben gesammelt, um eine ganzheitliche Analyse der

Determinanten der Risikomanagement-Entscheidungen durchzuführen. Drittens

werden Landwirte von agrartouristischen und nicht-agrartouristischen Betrieben

interviewt, um die zugrundeliegenden Motive des Einstiegs in den Agrartourismus

und Unterschiede in den Charakteristika der Betriebe zu untersuchen. Viertens

werden Daten von Landwirten im Ruhrgebiet analysiert um Erkenntnisse über

Determinanten der Diversifikationsentscheidung im peri-urbanen Raum zu

erhalten. Die gesammelten Daten werden mit Hilfe von ökonometrischen

Methoden analysiert.

Die Ergebnisse zeigen, dass die Kontextualisierung von experimentellen Methoden

der Risikopräferenzmessung Inkonsistenzen verringert und damit die Datenqualität

verbessert wird. Darüber hinaus zeigt die Analyse des

Risikomanagementverhaltens von Landwirten, dass der Entscheidungsrahmen des

einzelnen Landwirts (d.h. Risikopräferenzen, Risikowahrnehmung und

Erfahrungen mit Verlusten) die Wahl der Risikomanagementstrategie bestimmt.

Die Ergebnisse zeigen außerdem, dass die Nähe zu einem urbanen Agglomerat die

Landwirte dazu anregt, Diversifikationsaktivitäten zu betreiben. Die Entscheidung

zur nicht-landwirtschaftlichen Diversifizierung ist gelegenheitsbedingt, daher ist

eine bessere Information der Landwirte über bestehende und innovative

Diversifikationsmöglichkeiten für die ländliche Entwicklung essentiell.

Schlagwörter: Risikomanagement, experimentelle Methoden zur

Risikopräferenzmessung, Risikowahrnehmung,

Entscheidungsanalyse, nicht-landwirtschaftliche Diversifikation

i

Contents

Introduction ............................................................................................................. 7

1.1 Theoretical foundations of decision analysis under risk ........................ 10

1.1.1 Risk perception ............................................................................. 11

1.1.2 Risk preferences ............................................................................ 12

1.2 Research objective and structure of the thesis ....................................... 15

1.3.1 Research questions ........................................................................ 16

1.3.2 Structure of the thesis ................................................................... 21

1.3 Summary of main findings and conclusion ........................................... 22

1.4.1. Summary of main findings............................................................ 22

1.4.2. Conclusion and limitations ........................................................... 26

1.4 References ............................................................................................. 31

Using involvement to reduce inconsistencies in risk preference

elicitation .......................................................................................................... 36

2.1 Introduction ........................................................................................... 38

2.2 Literature background............................................................................ 40

2.3 Experimental design and methodology ................................................. 46

2.3.1 The self-assessment of risk preferences ........................................ 48

2.3.2 The iterative Multiple Price List ................................................. 48

2.4 Inconsistencies and contextualization.................................................... 53

2.5 Sample description and results .............................................................. 56

2.6 Conclusion ............................................................................................. 64

2.7 References ............................................................................................. 67

2.8 Appendix 2.A ........................................................................................ 71

2.9 Appendix 2.B ......................................................................................... 72

Risk perceptions, preferences and management strategies: Evidence

from a case study using German livestock farmers ...................................... 76

3.1 Introduction ........................................................................................... 78

3.2 Conceptual framework and relevant literature ...................................... 81

3.3 Experimental design .............................................................................. 86

3.4 Methodology .......................................................................................... 94

3.5 Data........................................................................................................ 96

3.6 Results ................................................................................................. 100

ii

3.7 Discussion ............................................................................................ 106

3.8 Conclusion ........................................................................................... 110

3.9 References ........................................................................................... 112

3.10 Appendix 3.A ...................................................................................... 117

Determinants and motives for agritourism activities: A German case

study ................................................................................................................ 124

4.1 Introduction ......................................................................................... 126

4.2 Theoretical framework ........................................................................ 128

4.2.1 Motives for agritourism initiation ............................................... 129

4.2.2 Role of farmwomen in the agritourism venture .......................... 131

4.2.3 Farm, farmer and household characteristics ............................... 132

4.2.4 Farmers’ risk preferences ............................................................ 134

4.3 Sampling and data ............................................................................... 135

4.3.1 Sampling and case study area ..................................................... 135

4.3.2 Data description .......................................................................... 138

4.4 Results ................................................................................................. 142

4.4.1 Motives behind agritourism ........................................................ 142

4.4.2 Role of farmwomen in agritourism ventures .............................. 146

4.4.3 Farm, farmer and household characteristics ............................... 147

4.5 Discussion and conclusion................................................................... 150

4.6 References ........................................................................................... 156

4.7 Appendix 4.A ...................................................................................... 161

Diversification in peri-urban agriculture: a case study in the Ruhr

metropolitan region ....................................................................................... 164

5.1 Introduction ......................................................................................... 166

5.2 Conceptual model and theoretical background.................................... 169

5.2.1 Diversification determinants ....................................................... 170

5.3 Data and case study area ...................................................................... 174

5.3.1 Case study area ........................................................................... 174

5.3.2 Data and data collection .............................................................. 176

5.4 Econometric model specification ........................................................ 182

5.5 Results and discussion ......................................................................... 184

5.6 Conclusion ........................................................................................... 190

5.7 References ........................................................................................... 192

iii

List of Tables

Table 2.1: Standard payoff table MPL ................................................................... 49

Table 2.2: Summary statistics by within-method consistency ............................... 57

Table 2.3: Summary statistics by frame and within method consistency .............. 58

Table 2.4: Contingency table of inconsistencies by task involvement................... 59

Table 2.5: Contingency table of inconsistencies by context involvement ............. 60

Table 2.6: Estimates of risk aversion and Luce structural noise parameter in

different iMPL frames ........................................................................... 61

Table 2.7: Spearman's rank correlations between CRRA interval mid-points in

different iMPL frames and self-assessment .......................................... 62

Table 2.8: Spearman's rank correlations between CRRA interval mid-points in

different iMPL frames and self-assessment for different levels of

task involvement ................................................................................... 63

Table 2.A.1: Variable description .......................................................................... 71

Table 3.1: Standard payoff table MPL ................................................................... 90

Table 3.2: Variable description .............................................................................. 98

Table 3.3: Summary statistics of complete surveys N = 56 ................................... 99

Table 3.4: Spearman correlation coefficients of risk preferences elicited with

different methods ................................................................................ 101

Table 3.5: Multinomial probit estimates for MPL, SA and average BS .............. 104

Table 3.6: Multinomial probit estimates for BS production, BS marketing and

prices, BS finances and BS agriculture generally ............................... 105

Table 3.A.1: Sources of risk included in the survey to elicit overall risk

perception scores ................................................................................. 117

Table 3.A.2: Risk management strategies included in the survey........................ 119

Table 3.A.3: Pearson correlation coefficients of risk preferences elicited with

different methods ................................................................................ 119

Table 3.A.4: Spearman correlation coefficients of risk preferences elicited

with different methods and risk perception in different domains ....... 121

Table 3.A.5: Multinomial logit estimates for MPL, SA and average BS............. 122

Table 3.A.6: Multinomial logit estimates for BS production, BS marketing

and prices, BS finances and BS agriculture generally ........................ 123

Table 4.1: Descriptive statistics of suggested motives for starting an

agritourism activity ............................................................................. 143

iv

Table 4.2: Summary statistic of agritourism ventures by type and farmwomen

participation ........................................................................................ 146

Table 4.3: Variable description ............................................................................ 149

Table 4.4: Summary statistics of farm farmer and household characteristics

for agritourism and non-agritourism farms ......................................... 150

Table 5.1: Description of explanatory variables .................................................. 181

Table 5.2: Summary statistics of explanatory variables ....................................... 182

Table 5.3: Estimated results for the double hurdle model .................................... 187

v

List of Figures

Figure 1.1: Farm resource allocation ....................................................................... 9

Figure 1.2: Factors affecting the farmers’ decision on resource allocation ........... 15

Figure 2.A.1: Distribution of the CRRA interval mid-point in two different

iMPL frames ......................................................................................... 72

Figure 2.A.2: Distribution of self-assessment task ................................................ 72

Figure 2.B.1: Third screen: control question ......................................................... 74

Figure 2.B.2: Fourth screen: General lottery task .................................................. 75

Figure 2.B.3: Fifth screen: Self-assessment task ................................................... 75

Figure 3.1: Factors influencing farmers’ decision-making on farm resource

allocation dimensions ............................................................................ 82

Figure 3.A.1: Example visual presentation MPL ................................................. 118

Figure 3.A.2: Kernel density plots of three different risk preference elicitation

methods ............................................................................................... 120

Figure 4.1: Location of case study area ............................................................... 138

Figure 4.2: Differences in motive categories ....................................................... 145

Figure 4.A.1: Multiple price list (MPL) with agricultural frame ......................... 161

Figure 5.1: Determinants influencing farmer’s diversification decision

adapted from van Raaij (1981) ............................................................ 170

Figure 5.2: Map of the Ruhr metropolitan area and the location of 132

surveyed farms .................................................................................... 176

vi

Abbreviations

AIC Akaike's Information Criterion

BS Business Statement

CAP Common Agricultural Policy

CRRA Constant Relative Risk Aversion

EU European Union

EUT Expected Utility Theory

EVA Expected Values A

EVB Expected Values B

iMPL iterative Multiple Price List

LR Likelihood Ratio

MPL Multiple Price List

SA Self-assessment

SEUT Subjective Expected Utility Theory

SOEP German Socio-Economic Panel

US United States

7

Chapter 1

Introduction

Agriculture is an inherently risky business. Risks faced by farmers are numerous, and

vary over space and time. Agricultural production is unique with regard to risk

exposure, as it is mostly performed outside, or includes live material. This makes

agriculture vulnerable to production risks such as extreme weather events, diseases and

infestations not found in other sectors (Hardaker et al., 2004, Huirne, Meuwissen, and

van Asseldonk, 2007, Anderson, Dillon, and Hardaker, 1977). Moreover, farmers are

exposed to market and price risks, prices for inputs like fertilizer, fodder services and

machinery, and outputs like milk, meat and cereal are not known a priori i.e. when

production decisions are made leading to volatile commodity prices in agriculture.

Additionally, price uncertainty is supported by open world markets, making

developments more unpredictable for single farmers. By introducing income

stabilization mechanisms like direct payments, guaranteed prices and buffer stocks

governments seek to mitigate some market risks. Policy makers also seek to regulate

numerous aspects of agriculture (e.g. the use of pesticides, animal welfare or land use

restrictions) leading to far reaching consequences for agricultural production.

Concurrently, high governmental support and regulations can lead to additional

uncertainty for farmers when proposed price stabilization mechanisms are abolished or

production regulations change. Thus, institutional risks refer to uncertainty associated

with changes in policies. Furthermore, farmers face financial risks that are related to

1.1 Theoretical foundations of decision analysis under risk

8

the financing of the farming business. Farms with a high debt-equity ratio are exposed

to fluctuating interest rates on borrowed capital, or have difficulties in making new

investments. Moreover, farmers face human or personal risks. They are common to all

business operators, nevertheless, are especially threatening for family businesses in the

agricultural sector. The death, illness or divorce of a family member or the main farm

operator can lead to substantial consequences for the farm business. In summary, risks

in agriculture are specific to geographical and climatic conditions, and depend on the

current market, political, financial and private business environment. Moreover the

farmers’ exposure to risks is time sensitive. For example production risks can change

due to increased catastrophic events such as floods and livestock epidemics caused by

climate change. Furthermore, western policy makers are pressured to shift away from

agricultural income support systems and publicly funded disaster programs, resulting

in greater exposure of farmers to competitive markets and private risk management

solutions (Meuwissen, van Asseldonk, and Huirne, 2008, Hardaker et al., 2004).

Consequently, agricultural risk management is gradually shifted back to the farmers’

decision making domain and hence gains importance for farmers, advisors, academics

and policy makers. The farmers’ choice of strategy to mitigate risks, i.e. the choice of

risk management strategy is in focus of this thesis.

When farmers decide to mitigate the risk they are exposed to, they trade some

of their expected returns for less variation in expected returns i.e. they shift parts of

their resources (e.g. land, capital and labor) away from traditionally risky agricultural

production towards a risk reducing strategy. Risk management strategies are


9

categorized in measures to i) share risks with others (i.e. off-farm strategies) or ii)

reduce risks within the farm (i.e. on-farm strategies) (Hardaker et al., 2004, Huirne,

Meuwissen, and van Asseldonk, 2007). Building upon farm diversification literature

(see e.g. van der Ploeg and Roep, 2003, Meraner et al., 2015), the latter is expanded by

differentiating between on-farm agricultural and on-farm non-agricultural risk

management strategies. The farms resource allocation under risk is illustrated in in

Figure 1.1.

Figure 1.1: Farm resource allocation

Farmers choose to mitigate agricultural risk by shifting resources away from

agricultural production towards three different dimensions. First, some risks can be

mitigated on the farm by adapting agricultural production (e.g. choice of breed,

investment in new technologies, combining different agricultural activities), and these

shall be henceforth referred to as on-farm agricultural strategies. Second, farmers can

keep resources on the farm but shift away from the agricultural production (e.g. non-

agricultural diversification, holding reserves or cutting private expenses) these

strategies are referred to as on-farm non-agricultural strategies. Third, risks are shared


10

with others, i.e. resources are shifted away from agricultural production off the farm

(i.e. off-farm strategies). This includes insurances, contracts with suppliers and buyers,

trading on futures markets and off-farm investments or employment.

Following this brief introduction on agricultural risks and risk management

strategies, theoretical foundations of decision making under risk, including risk

preferences and risk perception are introduced. Additionally, the research questions

addressed in the following chapters are presented. This first introductory chapter

concludes with a summary of the main findings as well as pointers on potential areas

for future research.


The various methods that have been developed for analyzing choices involving risk are

collectively called decision analysis (Hardaker et al., 2004). The theoretical

foundations of rational choice under uncertainty are commonly found in expected

utility theory (EUT) developed by von Neumann and Morgenstern (1947) based on

mathematical concepts established by Bernoulli (1738). In EUT, the rational decision

maker chooses between risky alternatives by comparing expected utility values of

different risky choices. Utility values refer to the weighted sums obtained by adding

the utility values of outcomes multiplied by the respective probabilities (Chavas, 2004,

Hardaker, 2006). EUT assumes that agents have perfect information on the

probabilities and potential consequences related to risky events. The model has been

criticized by numerous empirical studies as it fails to explain observable behavior (e.g.


11

Kahneman and Tversky, 1979, Savage, 1972, Allais, 1984). Savage (1972) expands

EUT by adding a subjective component to the expected utility hypothesis (i.e.

subjective expected utility (SEU)). SEU theory incorporates two components that have

to be assessed when analyzing decision makers’ choice under risk. This is, i) the

decision makers subjective beliefs about the probability and impact of an uncertain

outcome (i.e. subjective risk perception) as well as ii) the relative risk preferences for

uncertain outcomes (i.e. risk preferences) of the decision maker, evaluated via the

decision maker’s utility function (Hardaker, 2006).

1.1.1 Risk perception

Risk perception is commonly defined as the combination of the probability of

occurrence of an uncertain event and the consequential negative impact (Slovic,

Fischhoff, and Lichtenstein, 1982). Moreover, risk perception is dependent on the

objective risk the decision maker is facing, thus the perception of risk is domain-

specific (i.e. depending on the risk source).

There are numerous ways to measure subjective risk perception. Most

commonly, farmers’ risk perception is measured using multi-item Likert scale-type

questions or risk assessment scales of the two components of risk magnitude and

likelihood of risks (see e.g. van Winsen et al., 2014, Assefa, Meuwissen, and Oude

Lansink, 2016, Schaper, Bronsema, and Theuvsen, 2012). Furthermore, the visual

impact method (Hardaker et al., 2004) or the exchangeability method (Baillon, 2008,

Menapace, Colson, and Raffaelli, 2015a) have been used in the agricultural context.

The choice of method is dependent on the cognitive abilities of the decision maker.


12

More sophisticated methods run the risk of a lower comprehension and could therefore

lead to less meaningful results. Additionally, the choice of method is dependent on the

objective risk farmers are facing (e.g. multiple unknown sources of risk vs. one

dominating risk source).

1.1.2 Risk preferences

Regardless of their risk perception, different individuals have different attitudes

towards risk. Risk attitudes are referred to as willingness to take risks or risk

preferences. Choice under risk involves a trade-off between risk and expected return.

Risk averse decision makers prefer relatively low levels of risk. In other words, they

prefer to reduce the variations in possible outcomes and are in turn willing to sacrifice

some expected return. Risk seeking decision makers prefer relatively high levels of risk

and are willing to trade some expected return for a greater variation in possible

outcomes (March and Shapira, 1987). In EUT, the attitude towards risk is defined by

the shape of a utility function for money or wealth, with risk aversion depicted as

diminishing marginal utility, i.e. increasing at a decreasing rate (Hardaker et al., 2004,

Chavas, 2004).

Although the decision maker’s preferences for risk is assumed to be a partially

stable feature of individual personality, there is evidence that risk preferences are

influenced by variable factors such as mood, feelings, past experiences, ability to

comprehend probability statements and the way in which problems are framed

(Kahneman and Tversky, 1979, March and Shapira, 1987).


13

There are three major approaches to quantify risk attitudes: i) observed

economic behavior, e.g. econometric analysis (Antle, 1987, Just and Pope, 2002), ii)

non-incentivized questionnaires, e.g. Likert scales (Dohmen et al., 2011, Weber, Blais,

and Betz, 2002), and iii) experimental methods, e.g. lotteries or direct elicitation of the

utility function (Holt and Laury, 2002, Eckel and Grossman, 2002, Binswanger, 1981,

Pennings and Garcia, 2001). The primary, and biggest advantage of experimental and

survey based methods is the direct identification of otherwise latent variables, i.e. not

directly observable variables (Cox and Harrison, 2008). Secondly, risk preferences can

be elicited individually whereas field data used for econometric analysis is usually

available on an aggregated level. Thirdly, experimental methods provide the possibility

to control conditions for all participants, enabling researchers to analyze the decision

makers’ frame of decision making (Binswanger, 1981, Kahneman and Tversky, 1979).

Nevertheless, the choice of risk preference elicitation method is dependent on the

researchers resources (i.e. available data, time and budget) as well as on the decision

makers’ cognitive abilities (Charness, Gneezy, and Imas, 2013, Dave et al., 2010).

Additionally, more complex methods like lotteries can be misunderstood by the

participants provoking inconsistent behavior, leading to decreased data quality.

In the tradition of economic psychology, van Raaij (1981) is proposing a

conceptual framework including personal characteristics as well as risk perception to

explain economic behavior. Ilbery (1991) refers to “external” and “internal” drivers of

farmers’ decision making. More recently, van Winsen et al. (2014) propose a

framework including i) farming attitudes, (ii) socio-demographic characteristics, (iii)


14

past experiences, iv) risk attitude and (v) risk perception. By incorporating elements of

the presented literature, the following classification of factors influencing farmer’s

decision making is proposed: i) environment, ii) personal characteristics and iii) frame

of reference. The environment includes external farm characteristics (e.g. distance to

urban centers or natural protection sites, soil quality, institutional environment etc.).

Furthermore, personal characteristics are found to play a role in the choice of risk

management strategy (e.g. age, education, gender). Within the agricultural context of

decision analysis, characteristics specific to the agricultural business and household

(e.g. farm size, succession of the farm business, availability of family workforce and

the participation of the farm operator’s spouse) are henceforth synonymously referred

to as personal characteristics. Finally, the aforementioned subjective risk perception

and risk preferences are part of the decision makers’ frame of reference, and influence

the decision makers’ perceptual world, establishing the decision makers’ reality on

which decisions are based (Slovic, Fischhoff, and Lichtenstein, 1982, March and

Shapira, 1987).

1.2 Research objective and structure of the thesis

15

Figure 1.2: Factors affecting the farmers’ decision on resource allocation

Based on the decision environment, personal characteristics and frame of

reference, farmers choose the expected utility maximizing risk management portfolio

from a large set of available strategies. Consequentially the farmers’ choice of resource

allocation depicted in Figure 1.1 can be expanded by the factors affecting the farmers’

choice (see Figure 1.2).


The overall objective of this thesis is to investigate farmers’ choice of risk management

strategies in North-Rhine-Westphalia. Therefore, first contextualized experimental and

survey based risk preference elicitation methods are analyzed to identify an improved

application to the agricultural context. Second, determinants of the farmers’ choice of

risk management portfolio is empirically analyzed. Third, motives behind the farmers’


16

decision to diversify the farming activity towards agritourism are explored, and fourth

the farmers’ choice of on-farm non-agricultural diversification strategies is analyzed in

the peri-urban context. The underlying research questions, and their contribution to the

literature are presented in the following.

1.3.1 Research questions

To address the overarching objective of this thesis, four main research questions are

specified.

(I) Can contextualization of experimental risk preference elicitation methods reduce

inconsistencies?

The extent to which people are willing to take on risk constitutes their risk

attitude. Understanding individual attitudes towards risk is closely linked to the

goal of understanding and predicting economic behavior under uncertainty (like

the farmers’ choice of risk management strategy) and aiding policy-making.

Researchers have developed a variety of different experimental and survey based

methods to elicit individual risk preferences. Holt and Laury (2002) introduced

the most prominent experimental method to elicit risk preference (Anderson and

Mellor, 2009, Hellerstein, Higgins, and Horowitz, 2013). The presented lottery is

based on a multiple price list (MPL) format, where subjects successively pick the

option they prefer in a list of ten choices. Survey based methods traditionally

include multi-item, Likert-type scales. The simplest version is asking participants

to self-assess their general willingness to take risks on a scale from 0 to 10

(Dohmen et al., 2011).


17

The biggest disadvantage attached to experimental methods are inconsistencies in

the sample populations’ behavior. Here, two types of inconsistencies are

distinguished: The first type of within-method inconsistencies refer to a violation

of expected utility assumptions within a MPL setting. Second, between-method

inconsistencies describe non-consistent risk preferences of an individual when

different elicitation methods are used. Inconsistencies lead to a decrease of data

quality, generating biased real-world conclusions on human behavior and policy

recommendations. A large body of research aiming to overcome these issues

discusses new methods to elicit risk preferences (Crosetto and Filippin, 2013,

Dave et al., 2010, Lejuez et al., 2003) leading to a vast growth of experimental

methods to elicit risk preferences over the last decades without a solution to the

problem of inconsistent behavior. However, only little research has been done on

the impact of contextual adjustments of the classical abstract, context free lottery

setting. Here, contextualized lotteries refer to ones where subject specific real

world framing of the lottery task is employed. Abstract lotteries, on the other

hand, refer to ones where the instructions are kept abstract and context free

(Alekseev, Charness, and Gneezy, 2017). Contextualization is particularly

promising for field experiments dealing with participants not used to abstract

contexts like farmers. Therefore, chapter 2 of this thesis contributes to the

literature by comparing within-method inconsistencies in a controlled experiment

among agricultural students using a subject specific contextual framing of the

classical Holt and Laury (2002) lottery versus the original abstract and context


18

free frame. Additionally, the effect of contextualization of the MPL on between-

method inconsistency is analyzed by comparing risk preferences elicited using a

MPL and a general self-assessment task (following Dohmen et al., 2011).

(II) What determines the choice of agricultural risk management strategies?

In agricultural production, farmers are confronted with a wide and increasing

range of production, market, financial and institutional risks. Consequently, the

portfolio of risk management strategies available to farmers is large and growing,

but little is known about the determinants leading to the farmer’s choice of an

optimal risk management portfolio. The farmer’s choice of risk management

strategies is determined by farm, farmer and household characteristics. Within

the set of farmer characteristics, individual risk perception, risk preferences and

experienced past losses need to be considered (see Figure 1.2) (Slovic, Fischhoff,

and Lichtenstein, 1982, Hardaker et al., 2004).

Literature on farmers’ choice of risk management strategies is often focusing on

the adoption of single activities such as, insurances (Menapace, Colson, and

Raffaelli, 2015b, Finger and Lehmann, 2012). However, farmers use a large

portfolio of different risk management strategies in order to react to different risk

sources (Musser and Patrick, 2002, Meuwissen, Huirne, and Hardaker, 2001).

Thus, the study presented in chapter 3 aims to contribute to the existing literature

by analyzing the effect of farm, farmer and household characteristics on the

farmer’s choice of a large set of risk management strategies.

(III) What are underlying motives of the farmer’s choice to diversify into agritourism?


19

Farm diversification is of key relevance to the agricultural sector as it

contributes to rural development, and reduces agricultural income risks by

spreading the farm households’ sources of income. About one third of farmers in

the EU diversify into non-agricultural activities (European Parliamentary

Research Service, 2016), using their own assets (land, buildings, labor) outside

the core agricultural production to stabilize and/or increase their income. The

underlying theoretical framework of farmers’ resource allocation is introduced

in Figure 1.1. Additionally, farm diversification in general, and agritourism in

particular contributes to rural development and improves economic opportunities

and accessibility in disadvantaged rural regions (European Commission, 1990).

There is a large body of research focusing on observable farm, farmer and

household characteristics determining the farmers’ diversification decision (e.g.

Meraner et al., 2015, Ilbery et al., 1998, McNamara and Weiss, 2005).

Additionally, underlying push and pull motives of the farmers’ decision to start

an agritourism venture beyond observable characteristics are analyzed (e.g.

Barbieri, 2009, Hansson et al., 2013). Moreover, research is focusing on the

income stabilizing characteristics of agritourism, interpreting agritourism as a

farm risk management strategy (e.g. Kostov and Lingard, 2003, Meuwissen et

al., 2001). Furthermore, there is research focusing on the importance of the farm

operator’s spouse and underlying motives when analyzing drivers of agritourism

(e.g. Haugen and Vik, 2008, Ollenburg and Buckley, 2007, McGehee, Kim, and

Jennings, 2007, Hansson et al., 2013). The study presented in chapter 4 helps to


20

better understand the importance of push and pull motives, as well as the farm

households family structure in the decision making process. This provides

valuable information for policy makers and advisors, aiming to support rural

development by supporting agritourism activities.

(IV) What determines on-farm non-agricultural diversification in the peri-urban Ruhr

metropolitan area?

In peri-urban areas the farmer’s decision environment is characterized by

increased demand, short supply chains and direct marketing opportunities

leading to an increased uptake of farm diversification activities (Wilson, 2007,

Zasada et al., 2011, Heimlich and Barnard, 1992). However, farms in peri-urban

areas also face high opportunity costs for land and labor, as well as increased

public control (Monaco et al., 2017). Thus, analyzing farm diversification in the

peri-urban context is of particular interest. The case study region ‘Ruhr

Metropolis’ is especially interesting as it is the largest polycentric agglomeration

in Germany, and has the highest average population density in North-Rhine-

Westphalia.

Previous research on on-farm non-agricultural diversification has shown that due

to synergies between activities (e.g. between farm processing and sales

activities), farm diversification strategies are often combined (Meraner et al.,

2015, Haugen and Vik, 2008, Carter, 1998). Chapter 5 presents an analysis of

the farm environment, the farmers’ frame of reference (including risk

perception, risk preferences and past experiences), as well as personal


21

characteristics of the farm, farmer and farm household as illustrated in Figure

1.2. A two-step analysis is used to shed light on determinants of the farmers’

decision to engage in on-farm non-agricultural diversification and diversification

intensity.

1.3.2 Structure of the thesis

Chapters 2-5 constitute the main body of the thesis, addressing the research questions

and the objective of this thesis in independent analyses. Chapter 2 answers research

question (I) by analyzing results of a controlled experiment conducted online

comparing consistencies of responses in framed and standard lotteries. This chapter is

based on the paper currently under review titled “Using involvement to reduce

inconsistencies in risk preference elicitation”. The article in chapter 3 addresses

research question (II) by presenting a holistic analysis of the farmers’ choice of risk

management portfolio, determined by farmers’ risk preferences, perception and past

experiences among other farm, farmer and household characteristics. The

corresponding article is titled “Risk perceptions, preferences and management

strategies: Evidence from a case study using German livestock farmers”. The analysis

presented in chapter 4 answer research question (III) by including the main farm

operator and the spouses’ role in the farm management decision as well as expanding

determinants of decision making by underlying motives. Focus lies on the farmers’

decision to start an agritourism venture. The chapter contains the article:

“Determinants and motives for agritourism activities: A German case study”. Finally,

research question (IV) is addressed in chapter 5, which includes a two-step analysis of

1.3 Summary of main findings and conclusion

22

the farmers’ diversification decision and choice of diversification intensity in the peri-

urban context. It is based on the article titled: “Diversification intensity in peri-urban

areas: the Ruhr metropolitan region”.


For each of the articles presented in chapter 2-5, main findings are summarized in this

section. This is followed by conclusions, placing results into the broader context of the

overall research question and limitations of the analysis.

1.4.1. Summary of main findings

(I) Contextualization experimental risk preference elicitation methods reduces

inconsistent behavior

The analysis of data collected in an online experiment with 244 German

agricultural students to answer research question (I) reveals a reduction of

inconsistencies in the students’ responses when using a contextualized framing

of the classical lottery introduced by Holt and Laury (2002) (in the further

referred to as Holt and Laury lottery). Contextualization of the Holt and Laury

lottery reduces behavior violating assumptions of expected utility theory

repeatedly found for abstract Holt and Laury lottery frames. Furthermore, a

comparison of risk preferences elicited with the contextualized lottery and a

Likert scale general self-assessment task (following Dohmen et al., 2011) shows

a reduction of between-method inconsistencies. The agricultural context frame


23

increases the students, motivation and time spent on the task. Increased

motivation is in turn leading to less within- and between-method inconsistent

behavior. Adding a meaningful context specific to the addressed target group

(e.g. agricultural investment in the case of agricultural students) enhances

understanding of the task. This is of particular importance when tasks require

sophisticated reasoning. Hence, there is evidence that contextualization of risk

preference elicitation tasks can lead to better data quality and better explanatory

power of the elicited risk preferences.

(II) The choice of agricultural risk management strategies is depending on farm

farmer’s and household characteristics.

We use a holistic approach to analyze the farmers’ choice of risk management

strategies to answer research question (II), by including a broad list of risk

management strategies as well as personal characteristics of the farm farmer and

farm household. Based on the theoretical framework of subjective utility theory,

the farmer’s personal characteristics include elements of the farmer’s frame of

reference i.e. subjective risk perception of different risk sources, risk preferences

(comparing three different elicited methods) and experiences with major losses in

the past. The analysis is based on data collected via a self-administered paper

pencil survey among 64 livestock farmers in North-Rhine-Westphalia. The

survey includes a large set of risk management strategies that are in the latter

analysis grouped depending on the farm dimension where resources are shifted

towards (i.e. off-farm, on-farm agricultural, on-farm non-agricultural).


24

Furthermore, risk perception is measured using the average over two Likert-type

scales, quantifying the perceived likelihood and impact of uncertainty. Farmers’

risk preferences are elicited using three different methods in order to additionally

interpret differences in the predictive power of real life behavior for different

methods. The influence of farm, farmer and household characteristics on the

choice of risk management strategy is analyzed using a multinomial probit

model.

Generally, risk averse farmers are found to be more likely to choose on-farm risk

management strategies over off-farm strategies. More specifically, risk aversion

in the financial domain is linked to a larger probability of farmers choosing

primarily on-farm agricultural strategies. Risk aversion in general and, in other

domains is linked to a larger probability of farmers choosing primarily on-farm

non-agricultural strategies. This points to a strong need for domain specific risk

preference elicitation. Furthermore, results show that risk perception, age,

subjective numeracy, farm succession, farm size and the proportion of rented

land have a significant impact on farmers’ risk behavior.

(III) Farmers starting agritourism ventures are mainly driven by opportunities arising

from the vicinity to a city

The analysis of farmers’ motives to start agritourism to answer research question

(III) is based on data collected via semi-structured face-to-face interviews with

33 farmers in the surroundings of the city of Muenster in North-Rhine-

Westphalia. In total 24 motives underlying the farmers’ decision to start


25

agritourism are ranked by the interviewed farmers on a five point Likert scale

(ranging from 1 = “not important” to 5 = “very important”). The strongest

motive for all interviewed farmers is the desire to exploit new market

opportunities, followed by the joy to work in agritourism and a passion for

working with people. Least important motives are tax reasons and the wish to

create employment opportunities for the future generation. Pull motives are on

average more relevant for the farmers’ agritourism decision. This means that

farmers deciding to start agritourism on the farm are more opportunity, and less

necessity driven. Furthermore, the analysis highlights the importance of joined

work of the farming couple in the agritourism activity. Farmwomen are found to

be more often responsible for the realization of activities than the initialization

and planning. An additional comparison of characteristics on the farm, farmer

and household level of agritourism farms and non-agritourism farms in the same

area reveals that agritourism farms are smaller, more diversified and more often

run by female farm operators. Agritourism farm operators are also more risk

averse, indicating that more risk averse producers allocate more resources to

activities with less income volatility like agritourism.

(IV) On-farm non-agricultural diversification is an important risk management

strategy in the peri-urban Ruhr metropolitan area

In order to answer research question (IV) data from an online survey targeting

farmers in the peri-urban Ruhr metropolitan area is analyzed. The data is

enriched with geographical information on 156 participating farmers. A double


26

hurdle model is used to first analyze the farmers’ decision to shift resources

away from agricultural production towards on-farm non-agricultural

diversification and in a second step to analyze the intensity of diversification.

In the densely populated case study area, farms are to a large share diversified,

exploiting the advantages of the proximity to the urban agglomeration. The

farmers’ frame of reference plays an important role in determining the

diversification decision. Farmers that perceive market and price risks to be high,

and farmers that experienced severe losses due to institutional risks are less

likely to choose an on-farm non-agricultural diversification strategy. However,

risk aversion, high perception of labor risk and experienced losses due to market

risks decrease the farmers’ likelihood to diversify. Personal characteristics like

the farmer’s age, occupation and succession also influence the diversification

decision. More specifically, younger, full-time farming farmer with a secured

succession are more likely to seek income stabilizing on-farm non-agricultural

diversification. Diversification intensity is positively associated with farms that

are producing high value crops. These farm types are more likely to exploit

advantages of short supply chains, direct marketing opportunities and the

increased environmental awareness of consumers regarding agricultural

production in peri-urban areas.

1.4.2. Conclusion and limitations

With regard to the overall research objective of this thesis which can be described as an

investigation of the farmers’ choice of risk management strategies in North-Rhine-


27

Westphalia, the results can be split in two parts: the first one focusing on

improvements of experimental risk preference elicitation, and the second one on

empirical evidence for determinants of farmers risk management choices.

The analysis of contextualization effects on inconsistencies in experimental

risk preference elicitation elaborated on in research question (I), shows that context-

rich language adapted to the participants real world experiences in a controlled

experiment reduces inconsistencies. Thus, the commonly used argument for using an

abstract context in experimental economics to achieve experimental control is

challenged by the attained increase in participants’ understanding and motivation in a

contextualized experiment. Overall, the reduction of inconsistencies in the

contextualized experiment improves data quality. However, it is important to

acknowledge that contextualization of instructions may affect participants’ behavior in

the experiment which acts as a disadvantage. Nonetheless, this effect can be

appropriate as it relates to the research question and better explains real world behavior

in a subject-relevant context i.e. increasing external validity.

Besides the importance of risk preferences as determinant of the risk

management choices of livestock farmers in North-Rhine-Westphalia (research

question II), other elements of the farmers’ frame of reference, like risk perception,

risk literacy and experiences with severe losses are found to be significant. Farmers

perceiving market and price risks to be severe are more likely to choose risk sharing

strategies. Hence, a strong perception of volatile prices and insecure markets leads

farmers to choose more strategies targeting those risks. High risk aversion, low risk


28

literacy and experiences with severe losses in the past have a positive effect on the

uptake probability of on-farm risk management strategies as opposed to off-farm

strategies. In other words, farmers that seek to avoid risks, experienced severe losses in

the past and are less able to acquire and decode information on risk (i.e. compare and

transform probabilities and proportions) are less likely to share risks with third parties.

These farmers could profit most from additional support by advisors and policy makers

which need to tailor risk communication to the individual numeracy skills of farmers.

Additionally, the strong influence of the farmers’ frame of reference points to the need

for more in-depth insights into underlying motives and the role of other farm family

members in risk management choices. The analysis of farmers’ determinants to start

agritourism activities focuses on the farmers’ underlying motives (research question

III). Findings highlight that farmers choosing agritourism as a risk management

strategy in Muenster are opportunity driven. Interviewed farmers state that making use

of market opportunities emerging from the farms favorable location (e.g. proximity to

the city and recreational areas) enabled them to start the agritourism venture.

Furthermore, the generation of extra income, family structure (i.e. the participation of

farmwomen), and intrinsic motivation to work with people on the farm are important

factors. Findings indicate that the farms proximity to a city has a positive effect on the

uptake of agritourism. Yet, the comparison of agritourism and non-agritourism farms

in the same area shows that larger farms, with more risk seeking, male main farm

operators choose other risk management strategies over agritourism. These results

point towards the need to further investigate the role of the farms geographical


29

environment and market opportunities in peri-urban areas. Thus, when analyzing the

farmers’ choice of on-farm non-agricultural diversification strategies in the peri-urban

Ruhr metropolitan area characteristics of the farms geographical environment are

included (research question IV). On-farm non-agricultural diversification is a viable

farm risk management strategy in the peri-urban context. Nevertheless, not all farms

have the same conditions enabling them to diversify their activities. Farms that

produce high value crops, with a younger farmer and secured succession of the

business are more likely to take advantage of short supply chains and increasing local

customer demand, characteristic to peri-urban areas. Farm diversification is strongly

determined by the farmers’ frame of reference. Results show that past losses due to

changes in the political agenda and market volatility influence the farmers’ decision to

engage in on-farm non-agricultural diversification. This leads to the conclusion that the

current developments of the CAP towards supporting farmers to monetize

multifunctional characteristics of agriculture is fruitful. Furthermore, risk aversion has

a negative effect on the uptake probability of on-farm non-agricultural diversification

activities. We conclude that risk averse farmers prefer to shift the farming risk towards

third parties. Moreover, due to the already high degree of on-farm non-agricultural

diversification in the area for some activities saturation effects can play an important

role making the entrepreneurial decision to start an on-farm non-agricultural activity

more risky.

The empirical case studies included in this thesis show that farming systems

are very heterogeneous, offering different risk management opportunities to farmers,


30

while also shaping the individual frame of reference within which farmers are forming

their decisions. Policies should therefore be tailored to account for the special role of

these farming systems, with respect to, but also beyond diversification decisions.

Along these lines, when designing new policies like the income stabilization tool in

Europe, agricultural policy makers and advisers also need to recognize that risk and

risk aversion influence farmers' management decisions. Further research needs to

account for the large risk management portfolio available to farmers, while also

including risk preferences, perceptions, and past experiences in the decision making

analysis. The widespread analysis of factors influencing one single strategy ignores the

context of overall risk the farm is exposed to. Along these lines, future research needs

to account for the family business structure dominant in the agricultural sector.

Decisions in the agricultural risk management context are not made by the main farm

operator in isolation but are influenced by decisions of other family members, e.g.

spouse or potential successor. Consequently, their decision making frame of reference

plays a role in explaining farm risk management choices. Moreover, the analysis of

risk management strategies should consider a longer time horizon, enabling researchers

to dig deeper into the underlying determinants of farm risk management choices.

1.4 References

31

1.4 References

Alekseev, Aleksandr, Gary Charness, and Uri Gneezy. 2017. "Experimental methods:

When and why contextual instructions are important." Journal of Economic

Behavior & Organization 134:48-59.

Allais, Maurice. 1984. "The Foundations of the Theory of Utility and Risk Some

Central Points of The Discussions at the Oslo Conference." In Progress in Utility

and Risk Theory, edited by Ole Hagen and Fred Wenstop, 3-131. Dordrecht NL,

Boston USA, Lancaster UK: D. Reidel Publishing Company.

Anderson, Jock R., John L. Dillon, and Brian J. Hardaker. 1977. Agricultural decision

analysis. Iowa City USA: Iowa State University Press.

Anderson, Lisa R., and Jennifer M. Mellor. 2009. "Are risk preferences stable?

Comparing an experimental measure with a validated survey-based measure."

Journal of Risk and Uncertainty 39 (2):137-160.

Antle, John M. 1987. "Econometric estimation of producers' risk attitudes." American

Journal of Agricultural Economics 69 (3):509-522.

Assefa, Tsion T., Miranda P. M. Meuwissen, and Alfons G. J. M. Oude Lansink. 2016.

"Price risk perceptions and management strategies in selected European food supply

chains: An exploratory approach." NJAS - Wageningen Journal of Life Sciences.

Baillon, Aurélien. 2008. "Eliciting subjective probabilities through exchangeable

events: An advantage and a limitation." Decision Analysis 5 (2):76-87.

Barbieri, Carla. 2009. "A comparison of agritourism and other farm entrepreneurs:

Implications for future tourism and sociological research on agritourism."

Northeastern Recreation Research Symposium, Bolton Landing, NY, USA, March

2008.

Bernoulli, Daniel. 1738. Specimen theoriae novae de mensura sortis. Vol. 5,

Commentarii Academiae Scientiarum Imperialis Petropolitae: Gregg.

Binswanger, Hans P. 1981. "Attitudes toward risk: Theoretical implications of an

experiment in rural India." The Economic Journal 91 (364):867-890.

Carter, Sara. 1998. "Portfolio entrepreneurship in the farm sector: indigenous growth in

rural areas?" Entrepreneurship & Regional Development 10 (1):17-32.

Charness, Gary, Uri Gneezy, and Alex Imas. 2013. "Experimental methods: Eliciting

risk preferences." Journal of Economic Behavior & Organization 87:43–51.

1.4 References

32

Chavas, Jean-Paul. 2004. Risk analysis in theory and practice. San Diego USA,

London UK: Elsevier Academic Press.

Cox, James C., and Glenn W. Harrison. 2008. "Risk aversion in experiments: An

introduction." In Risk aversion in experiments, edited by James C. Cox and Glenn

W. Harrison, 1-7. online: 08 Mar 2015: Emerald Group Publishing Limited.

Crosetto, Paolo, and Antonio Filippin. 2013. "The “bomb” risk elicitation task."


Dave, Chetan, Catherine C. Eckel, Cathleen A. Johnson, and Christian Rojas. 2010.

"Eliciting risk preferences: When is simple better?" Journal of Risk and

Uncertainty 41 (3):219-243.

Dohmen, Thomas, Armin Falk, David Huffman, Uwe Sunde, Jürgen Schupp, and Gert

G. Wagner. 2011. "Individual risk attitudes: Measurement, determinants, and

behavioral consequences." Journal of the European Economic Association 9

(3):522–550.

Eckel, Catherine C., and Philip J. Grossman. 2002. "Sex differences and statistical

stereotyping in attitudes toward financial risk." Evolution and Human Behavior 23

(4):281–295.

European Commission. 1990. Community Action to Promote Rural Tourism -

Communication from the Commission.

European Parliamentary Research Service. 2016. Farm diversification in the EU.

European Parliament: Briefing.

Finger, Robert, and Niklaus Lehmann. 2012. "The influence of direct payments on

farmers’ hail insurance decisions." Agricultural Economics 43 (3):343-354.

Hansson, Helena, Richard Ferguson, Christer Olofsson, and Leena Rantamäki-

Lahtinen. 2013. "Farmers' motives for diversifying their farm business – The

influence of family." Journal of Rural Studies 32:240-250.

Hardaker, Brian J. 2006. "Farm risk management: past, present and prospects."

Journal of Farm Management 12 (10):593-612.

Hardaker, Brian J., Ruud B. M. Huirne, Jock R. Anderson, and Gudbrand Lien. 2004.

Coping with risk in agriculture. 2 ed. Wallingford UK, Cambridge USA: CABI

publishing.

1.4 References

33

Haugen, Marit S., and Jostein Vik. 2008. "Farmers as entrepreneurs: the case of farm-

based tourism." International Journal of Entrepreneurship and Small Business 6

(3):321-336.

Heimlich, Ralph E., and Charles H. Barnard. 1992. "Agricultural adaptation to

urbanization: Farm types in northeast metropolitan areas." Northeastern Journal of

Agricultural and Resource Economics 21 (1):50-60.

Hellerstein, Daniel, Nathaniel Higgins, and John Horowitz. 2013. "The predictive

power of risk preference measures for farming decisions." European Review of

Agricultural Economics 40 (5):807–833.

Holt, Charles A., and Susan K. Laury. 2002. "Risk aversion and incentive effects."

American Economic Review 92 (5):1644–1655.

Huirne, Ruud B. M., Miranda P. M. Meuwissen, and Marcel A. P. M. van Asseldonk.

2007. "Importance of whole-farm risk management in agriculture." In Handbook of

operations research in natural resources, edited by Andres Weintraub, Carlos

Romero, Trond Bjørndal and Rafael Epstein, 3-15. New York USA: Springer US.

Ilbery, Brian W. 1991. "Farm diversification as an adjustment strategy on the urban

fringe of the West Midlands." Journal of Rural Studies 7 (3):207–218.

Ilbery, Brian W., Ian Bowler, Gordon Clark, Alastair Crockett, and Alastair Shaw.

1998. "Farmbased tourism as an alternative farm enterprise: A case study from the

northern Pennines, England." Regional Studies 32 (4):355–365.

Just, Richard E., and Rulon D. Pope, eds. 2002. A Comprehensive Assessment of the

Role of Risk in U.S. Agriculture. 1 ed, Natural Resource Management and Policy.

New York USA: Springer.

Kahneman, Daniel, and Amos Tversky. 1979. "Prospect theory: An analysis of

decision under risk." Econometrica: Journal of the Econometric Society:263-291.

Kostov, Philip, and John Lingard. 2003. "Risk management: a general framework for

rural development." Journal of Rural Studies 19 (4):463-476.

Lejuez, Carl W., Will M. Aklin, Michael J. Zvolensky, and Christina M. Pedulla. 2003.

"Evaluation of the Balloon Analogue Risk Task (BART) as a predictor of

adolescent real-world risk-taking behaviours." Journal of Adolescence 26 (4):475-

479.

March, James G., and Zur Shapira. 1987. "Managerial perspectives on risk and risk

taking." Management Science 33 (11):1404-1418.

1.4 References

34

McGehee, Nancy G., Kyungmi Kim, and Gayle R. Jennings. 2007. "Gender and

motivation for agri-tourism entrepreneurship." Tourism Management 28 (1):280–

289.

McNamara, Kevin T., and Christoph Weiss. 2005. "Farm Household Income and On-

and Off-Farm Diversification." Journal of Agricultural and Applied Economics 37

(1):37-48.

Menapace, Luisa, Gregory Colson, and Roberta Raffaelli. 2015a. "Climate change

beliefs and perceptions of agricultural risks: An application of the exchangeability

method." Global Environmental Change 35:70-81.

Menapace, Luisa, Gregory Colson, and Roberta Raffaelli. 2015b. "A comparison of

hypothetical risk attitude elicitation instruments for explaining farmer crop

insurance purchases." European Review of Agricultural Economics 43 (1):113-135.

Meraner, Manuela, Wim Heijman, Tom Kuhlman, and Robert Finger. 2015.

"Determinants of farm diversification in the Netherlands." Land Use Policy

42:767-780.

Meuwissen, Miranda P. M., Brian J. Hardaker, Rudd B. M. Huirne, and Aalt A.

Dijkhuizen. 2001. "Sharing risks in agriculture; principles and empirical results."

NJAS - Wageningen Journal of Life Sciences 49 (4):343–356.

Meuwissen, Miranda P. M., Rudd B. M. Huirne, and Brian J. Hardaker. 2001. "Risk

and risk management: an empirical analysis of Dutch livestock farmers." Livestock

Production Science 69 (1):43–53.

Meuwissen, Miranda P. M., Marcel A. P. M van Asseldonk, and Ruud B. M. Huirne.

2008. Income Stabilisation in European Agriculture: Design and Economic Impact

of Risk Management Tools. Wageningen NL: Wageningen Academic Publishers.

Monaco, Federica, Ingo Zasada, Dirk Wascher, Matjaž Glavan, Marina Pintar, Ulrich

Schmutz, Chiara Mazzocchi, Stefano Corsi, and Guido Sali. 2017. "Food

Production and Consumption: City Regions between Localism, Agricultural Land

Displacement, and Economic Competitiveness." Sustainability 9 (1):96.

Musser, Wesley N., and George F. Patrick. 2002. "How Much does Risk Really Matter

to Farmers?" In A Comprehensive Assessment of the Role of Risk in U.S.

Agriculture, edited by Richard E. Just and Rulon D. Pope, 537-556. Boston USA:

Springer US.

Ollenburg, Claudia, and Ralf Buckley. 2007. "Stated economic and social motivations

of farm tourism operators." Journal of Travel Research 45 (4):444-452.

1.4 References

35

Pennings, Joost M. E., and Philip Garcia. 2001. "Measuring producers' risk

preferences: A global risk-attitude construct." American Journal of Agricultural

Economics 83 (4):993-1009.

Savage, Leonard J. 1972. The Foundations of Statistics. New York USA: Courier

Corporation.

Schaper, Christian, Hauke Bronsema, and Ludwig Theuvsen. 2012. Risikomanagement

in der Landwirtschaft. Schriftenreihe des Landesamtes für Umwelt, Landwirtschaft

und Geologie 36, 36.

Slovic, Paul, Baruch Fischhoff, and Sarah Lichtenstein. 1982. "Why Study Risk

Perception?" Risk Analysis 2 (2):83–93.

van der Ploeg, Jan Douwe, and Dirk Roep. 2003. "Multifunctionality and rural

development: the actual situation in Europe." In Multifunctional agriculture: a new

paradigm for European agriculture and rural development, edited by Guido van

Huylenbroeck and Guy Durand, 37-53. Hampshire UK: Ashgate.

van Raaij, Fred W. 1981. "Economic psychology." Journal of Economic Psychology 1

(1):1–24.

van Winsen, Frankwin, Yann de Mey, Ludwig Lauwers, Steven van Passel, Mark

Vancauteren, and Erwin Wauters. 2014. "Determinants of risk behaviour: effects of

perceived risks and risk attitude on farmer’s adoption of risk management

strategies." Journal of Risk Research 19 (1):56-78.

von Neumann, John, and Oskar Morgenstern. 1947. Theory of games and economic

behavior. 2 ed. Princeton USA, Woodstock UK: Princeton University Press.

Weber, Elke U., Ann‐Renee Blais, and Nancy E. Betz. 2002. "A domain‐specific risk‐attitude scale: Measuring risk perceptions and risk behaviors." Journal of

Behavioral Decision Making 15 (4):263–290.

Wilson, Geoffrey Alan. 2007. Multifunctional agriculture: a transition theory

perspective. Wallingford UK, Cambridge USA: CABI publishing.

Zasada, Ingo, Christian Fertner, Annette Piorr, and Thomas Sick Nielsen. 2011. "Peri-

urbanisation and multifunctional adaptation of agriculture around Copenhagen."

Geografisk Tidsskrift-Danish Journal of Geography 111 (1):59–72.

36

Chapter 2

Using involvement to reduce inconsistencies in risk

preference elicitation*

* Meraner, M., O. Musshoff and R. Finger (under review)

37

Abstract

Empirical research aiming to elicit risk attitudes faces problems of within-

and between-method inconsistencies, which reduce the explanatory and

predictive power of risk research. In this paper, we examine the particular

relevance of context and task involvement on the elicitation of risk

preferences and inconsistencies. We find in a sample of 244 German

agricultural students, that a real life and subjects’ context specific framing of

a multiple price list triggers the participants’ motivation. Higher motivation

is in turn triggering analytic/systematic thinking and is leading to fewer

within- and between-method inconsistencies. We show that within-method

consistency is increased with subjects’ increasing task and context

involvement. However, between-method consistency is significantly

increased by the subjects’ task involvement; context involvement, has a less

distinct effect on between-method consistency. We show that by framing a

risk elicitation method according to the subjects’ specific context,

involvement can be triggered and inconsistencies and misspecifications can

be reduced.

Keywords: between- and within-method inconsistencies, risk

preference elicitation, involvement

JEL Classifications: C91 D81

2.1 Introduction 38

38

2.1 Introduction

The extent to which people are willing to take on risk constitutes their risk attitudes,

which in turn plays a major role in explaining their behavior. Consequentially risk

attitudes are of high importance for decisions in many economics-related contexts.

Understanding individual attitudes towards risk is closely linked to the goal of

understanding and predicting economic behavior and giving policy advice.

There is a growing literature on how to measure risk attitudes and accordingly

a large body of literature focuses on the selection of the right elicitation method (for an

extensive overview see Charness, Gneezy, and Imas, 2013). Many of these methods

are based on the same theoretical foundation of expected utility theory (EUT) and thus

claim to measure the subjects’ “true” risk preference. Consequently, risk preferences

elicited using different methods should be comparable and accurate. However, because

of inconsistencies (i.e. errors) in the individuals’ responses these criteria are often not

met in empirical work by the participants (Csermely and Rabas, 2017). More

specifically, three ways of consistency are distinguished in the literature i) between-

method consistency of several elicitation methods (Crosetto and Filippin, 2015,

Dohmen et al., 2011, Eckel and Grossman, 2002), ii) within-method consistency of the

same elicitation method at one point in time (Holt and Laury, 2002, Jacobson and

Petrie, 2009), and iii) within-method consistency of the same elicitation method over

two points in time (Andersen et al., 2008, Harrison and Rutström, 2008).

Inconsistencies lead to biases in the interpretation of the decision makers’ risk

2.1 Introduction 39

39

preferences and consequently biased real-world conclusions on human behavior and

policy recommendations. In order to overcome these inconsistency problems, past

research has frequently reached out to new methods to elicit risk preferences

(Charness, Gneezy, and Imas, 2013). This did not necessarily result in lower

inconsistencies but contributed to increasing problems of comparability of the different

studies. Furthermore, a large body of literature seeks to identify the correct

assumptions about the nature of the data gathered and thus ‘errors’ made by the

subjects in the experiments generating the data under analysis (Carbone and Hey,

2000, Wilcox, 2008).

Based on the seminal work of Kahneman and Tversky (1979), numerous

studies have shown that decision making is strongly influenced by the decision frame

(i.e. decision makers respond differently to different but objectively equivalent

descriptions of the same problem). Furthermore, Kahneman (2003) defined two

different ways of processing information applied in different contexts of decision

making depending on the motivation and capability of the decision maker. The

motivation of subjects is expected to be dependent on the subject-specific relevance of

the task, or, in other words, the subjects’ involvement with it. The subject-specific

relevance is expected to be influenced by the decision frame, so that framing can

trigger task involvement.

And indeed, there is evidence in different experimental settings that the

application of context is enhancing understanding of experimental tasks, reduces

mistakes and increases quality of results (see Alekseev, Charness, and Gneezy, 2017

2.2 Literature background 40

40

for an extensive overview). However, there is evidence of heterogeneity with respect to

how people respond to contextual changes. Alatas et al. (2009) conclude that expert

subjects find contextual framing more useful than students, leading to the conclusion

that framing effects in student subject pools might have been underestimated so far.

We aim to close the gap in the literature and to reduce inconsistencies by

including contextual framing and personal involvement in the risk elicitation research

design. More specifically, we show in this paper that risk preference elicitation

methods evoke fewer between-method and within-method inconsistencies when

specific task and context involvement is included in the analysis. In our analysis, task

involvement is determined by the decision makers’ task related effort. In contrast,

context involvement is defined by the personal relevance of the task for the decision

maker.

The remainder of this paper is organized as follows. First, we give an overview

on the existing literature concerning risk preference elicitation and inconsistencies.

Next, an introduction to the experimental design and methodology used in this analysis

is presented. The subsequent description of the data sample and results of this research

is followed by the conclusion.

2.2 Literature background

Over the last decade approximately 20 new methods to elicit risk preferences have

been published (for a detailed overview on the most established ones see Charness,

Gneezy, and Imas, 2013). Along these lines, there is growing literature on comparing


41

experimental methods to measure risk preferences (e.g. Coppola, 2014, Csermely and

Rabas, 2017, Crosetto and Filippin, 2015).

The most prevalent method to elicit risk preferences is via a Multiple Price List

(MPL), where subjects are presented with a series of choices between gambles. This

approach allows to estimate intervals for the curvature parameters of a utility function

for each subject. However, since the inference of risk preferences, and in turn,

parameter estimation, requires a unique switching point respondents with more than

one switching point are not behaving consistently under standard EUT assumptions on

preferences (Charness et al. 2013). The problem of inconsistencies in MPL tasks is

highly relevant in empirical research on experimental risk preference elicitation

methods. For instance, Charness and Viceisza (2016) found that 75% of Senegalese

farmers made inconsistent choices, Hirschauer et al. (2014) found 57% inconsistent

answers amongst Kazakh farmers, and, using a sample of adults in Ruanda, Jacobson

and Petrie (2009) found an inconsistency rate of 55%. High inconsistency rates are also

observed in developed countries: e.g. Lévy-Garboua et al. (2012) find that on average

around 36% of French students behave inconsistently in different MPL settings, Holt

and Laury (2002) find 13% inconsistent answers amongst students in the USA and

Dave et al. (2010) find 8.5% of participants answering inconsistently in a sample of

Canadian citizens. The main problem of data containing inconsistencies is related to

the different ways of dealing with inconsistencies to interpret risk preferences. Most

researchers choose to either ignore subjects with inconsistent choices or to make

specialized assumptions on the nature of stochastic errors and estimate the parameters


42

of interest (Jacobson and Petrie, 2009). Excluding inconsistently behaving subjects,

results in a biased sample since systematic differences may exist in the risk preferences

of consistent and inconsistent participants (see Jacobson and Petrie (2009) for more

details on behavioral patterns of subjects making mistakes). When including

inconsistently responding subjects in the estimation of the risk aversion parameter, a

stochastic error term (i.e. ‘structural noise’) parameter is often included in the

estimation (see e.g. Harrison and Rutström, 2008, Carbone and Hey, 2000).

Three driving factors explaining between- and within-method instability of risk

preference elicitation have been identified in the literature: i) differences in the

cognitive ability of subjects and task complexity (Lévy-Garboua et al., 2012, Anderson

and Mellor, 2009, Dave et al., 2010), ii) misspecification of individual preferences

(Harrison, Lau, and Rutström, 2007, Starmer, 2000) and iii) context-dependence of risk

preferences (Holt and Laury, 2005, 2002, Deck, Lee, and Reyes, 2014).

One way to overcome problems with inconsistencies stemming from the

subject pools cognitive abilities or complexity of the task is to use simpler risk

preference elicitation methods. Dave et al. (2010) perform experiments on subjects

with different mathematical ability. They conclude that a simpler elicitation method

results in higher within-method consistency for subjects with lower mathematical

ability. However, simpler alternative risk elicitation methods imply a loss of

comparability and accuracy. Furthermore, Bruner (2009) and Lévy-Garboua et al.

(2012) explore how different ways of displaying the choice sets affect inconsistency

rates. Bruner (2009) finds less within-method inconsistencies for a menu displayed


43

lottery frame with increasing probabilities vs. increasing reward. Levin, Schneider, and

Gaeth (1998) find more inconsistent behavior with a sequential and increasing

presentation of the MPL compared to a random probability presentation. Based on

these results, Lévy-Garboua et al. (2012) conclude that inconsistencies with a bad

frame, in terms of visual presentation of the MPL, are driven by a lack of information.

In a similar vein, Andersen et al. (2008) find cognitively more challenging tasks (risk

preference vs. time preference elicitation), to induce more noise in the estimated

parameter.

To overcome inconsistencies, due to misspecifications in the underlying

theoretical model. Some include elements of prospect theory e.g. loss aversion and

probability weighting to characterize risk attitudes (for a detailed comparison of

different underlying theoretical concepts see Abdellaoui, Driouchi, and L’Haridon,

2011). Other authors interpret inconsistencies as indifferences and hence adapt the

original design of the MPL i) by including a third choice in each row indicating

indifference in preference between both lotteries (Andersen et al., 2008) or ii) by

enforcing a unique switching point (see Harrison, Lau, and Rutström, 2007, Andersen

et al., 2006). The latter imposes strict monotonicity on revealed preferences and

enforces transitivity. As there is no further control mechanism to ascertain whether all

participants understood the task, this might cause biases of the results and, in turn,

biases of the estimated preferences.

Moreover, inconsistencies have been found to be context and stake dependent.

For instance, Holt and Laury (2002); (2005) find that inconsistencies can be reduced


44

by increasing the payoff level. The importance of the effect of decision frames on risk

preferences has been widely recognized in the literature on decision making analysis

(Levin, Schneider, and Gaeth, 1998, Tversky and Kahneman, 1986). Specifically,

Deck, Lee, and Reyes (2014) find that fewer inconsistencies occur if the MPL is

framed as financial investment task compared to a lottery task. They, however, used a

very general setting without accounting for the specific background of the participants.

Thus, we aim to extend the existing literature by focusing on the role of the subjects’

contextual and task involvement when analyzing inconsistencies and the effects of

different decision frames.

Based on McElroy and Seta (2003), we define task involvement as the

personal effort, motivation and capacity to perform the task at hand (we use the time

spent on a specific task as proxy). Context involvement is defined as the personal

relevance of the task for the decision maker (we use an involvement score based on the

student’s involvement with the agricultural domain to measure context involvement).

McElroy and Seta (2003) assume increasing task involvement with increasing context

involvement, or more specifically the motivation and capacity to solve a problem is

expected to increase with increasing personal relevance of the problem at hand.

Furthermore, they differentiate two ways of processing decision problems2 arising

from different levels of task and context involvement of the decision maker. In

particular, McElroy and Seta (2003) find that holistic/heuristic processing occurs with

low levels of motivation or capacity to solve a problem. Analytic/systematic thinking

2 Their approach follows lines of earlier research see e.g. Stanovich and West (1999) and

Tversky and Kahneman (1986).


45

in turn sets in when the subject’s motivation and ability are high3. Moreover, they

conclude that with increasing relevance of the decision, the amount of effort expended

on the task increases as well as the likelihood of analytic/systematic thinking. Subjects

with a higher likelihood of exhibiting an analytic/systematic processing style are found

to be more insensitive to the influence of framing effects.

However, this finding has not yet been considered in the approaches to reduce

inconsistent behavior. Thus, we hypothesize that this is even more relevant if the

investment task is placed in a subject specific setting. More specifically, we focus on

agricultural students and formulate the financial investment task as agricultural

investment decision to be taken by the participants. Furthermore, we measure the

degree of context involvement in agricultural activities for each participant and test the

influence of context involvement on consistency. Given the relevance of both between-

method and within-method consistency (Csermely and Rabas, 2017), we aim to

address both problems in our analysis. To this end, we consider “within-method

consistency” i.e. consistent behavior within the MPL at one point in time and

“between-method consistency” i.e. consistent behavior in different elicitation methods.

We use the subject’s task involvement (i.e. time spent on the risk elicitation task) to

test the influence on within-method inconsistencies. We hypothesize that both: within-

and between method consistency increases with increasing task and context

involvement.

3 Kahneman (2003) uses the terms system 1 thinking and system 2 thinking.

2.3 Experimental design and methodology 46

46

Furthermore, we use two different frames of the MPL to test the influence of

contextualization on the consistent behavior between different risk preference

elicitation methods (MPL and self-assessment). Following earlier studies e.g.

Anderson and Mellor (2009), Charness and Viceisza (2016), Dohmen et al. (2011) and

Thoma (2015), we compare risk preferences elicited using incentivized methods

(MPL) and not incentivized methods (self-assessment).

2.3 Experimental design and methodology

We focus on a homogeneous sample of students to reduce the influence of factors not

controlled for in the study. More specially, we conducted the survey with agricultural

science students at the two largest agricultural departments in the state of North-Rhine

Westphalia (Germany), i.e. the University of Bonn and the South Westphalian

University of Applied Sciences4. All students in agricultural sciences in both

universities were invited to participate in two identical online surveys conducted in

January and March 2015, respectively. We aligned the list of participating students

with the university administrative offices’ database, ensuring that the individual

student e-mail address used to complete the survey corresponds to students enrolled in

agricultural studies. Students participating without being enrolled in agricultural

studies at the time the survey was realized are excluded from further analysis5. The

experiment was conducted in two parts. Part I consisted of two risk-aversion tasks as

4 About 1,100 and 500 students are enrolled in agricultural sciences are enrolled in both

faculties, respectively. 5 There is no information on students who did not select to participate but we expect selection

biases to be small because we targeted a very homogenous group of students.


47

explained in more detail in the following and part II consisted of a questionnaire

collecting subjects’ socio-demographic characteristics. Specifically, we collected

information on age, sex, optimism and mothers’ highest educational degree and risk

literacy. For the latter, we used the Berlin numeracy test described in Cokely et al.

(2012). Additionally we included in this section specific characteristics to measure the

students’ context involvement (i.e. growing up on a farm holding, parents are farmers,

planned succession of a farm, type and length of specific agricultural education). We

also tested the effects of these characteristics on risk aversion, but do not find

significant effects, which is in line with similar research (e.g. Deck, Lee, and Reyes,

2014). This inexistence of significant effects of participants’ characteristics is expected

to be caused by the very homogenous sample used in this study. The results are not

presented here but are available upon request from the authors. We measured the time

each participant spent on each part of the questionnaire and use the time spent on the

iterative Multiple Price List (iMPL) as proxy for task involvement.

To elicit risk preferences we use two methods dominant in the literature: a self-

assessment of general risk preferences, and an iterative Multiple Price List (iMPL), an

extension of the MPL. We include two different decision frames in our experiment, i.e.

two different wordings that change the contextual setting of the iMPL. Additionally,

we randomly changed the order of the two risk preference elicitation methods (self-

assessment and iMPL). By using a random design assigning each participant only one

frame, we aim to control for potential biases arising from the sequence of tasks. The


48

instructions to the risk elicitation tasks presented to the subjects are available in

Appendix 2.B6.

2.3.1 The self-assessment of risk preferences

The self-assessment of general risk preferences is consistent with several other

researchers’ applications a 11-point Likert scale (Thoma, 2015, Dohmen et al., 2011,

Charness and Viceisza, 2016). The exact wording is taken from the German Socio-

Economic Panel Study (SOEP), which is (translated from German) as follows: “How

do you see yourself: are you generally a person who is fully prepared to take risks or

do you try to avoid taking risks? Please tick a box on the scale, where the value 0

means: `not at all willing to take risks' and the value 10 means: `very willing to take

risks'." (Wagner, Frick, and Schupp, 2007).

2.3.2 The iterative Multiple Price List

The iterative Multiple Price List (iMPL) is an extension of the basic MPL developed

by Harrison, Lau, and Rutström (2007). It elicits risk preferences, resulting in a more

refined description of the subjects risk preferences compared to the standard MPL. The

standard MPL as introduced by Holt and Laury (2002) is structured as follows: The

table has ten rows and two columns; in each row the subjects face two gambling

choices A and B. Table 2.1 illustrates the basic payoff matrix presented to the subjects.

Note that only the left side of the table is shown to the participants (i.e. not showing

6 Further insights on the data are available in the Data in Brief paper accompanying this article

(Meraner, Musshoff and Finger, submitted), the full survey is available in German upon

request.


49

the expected value of option A and B, the difference of expected values and the

constant relative risk aversion (CRRA) interval). The subjects are asked to choose

either A or B in each row. The intuition behind this test for risk aversion is that only

very risk-loving subjects would take option B in the first row, and only very risk-

averse subjects would take option A in the second-last row. The last row has no

relevance for risk aversion, and is simply a test that the subject understood the

instructions. Only choosing option B is a rational choice, independent of the level of

risk aversion. A risk-neutral subject should switch from choosing A to B when the

difference of expected values A (EVA) and B (EVB) is the smallest (see right side of

Table 2.1), so a risk-neutral subject would choose A for the first four rows and B

thereafter.

Table 2.1: Standard payoff table MPL

Option A Option B EVAa)

EVBa)

Differencea)

Open CRRA

intervala) b)

p(40€) p(32€) p(77€) p(2€)

10% 90% 10% 90% 32.80 9.50 23.30 r < -1.71

20% 80% 20% 80% 33.60 17.00 16.60 -1.71 < r ≤ -0.95

30% 70% 30% 70% 34.40 24.50 9.90 -0.96 < r ≤ -0.49

40% 60% 40% 60% 35.20 32.00 3.20 -0.50 < r ≤ -0.14

50% 50% 50% 50% 36.00 39.50 -3.50 -0.15 < r ≤ 0.15

60% 40% 60% 40% 36.80 47.00 -10.20 0.16 < r ≤ 0.41

70% 30% 70% 30% 37.60 54.50 -16.90 0.42 < r ≤ 0.68

80% 20% 80% 20% 38.40 62.00 -23.60 0.69 < r ≤ 0.97

90% 10% 90% 10% 39.20 69.50 -30.30 0.98 < r ≤ 1.37

100% 0% 100% 0% 40.00 77.00 -37.00 r > 1.37

Source: Own depiction according to Holt and Laury (2002) and Harrison, Lau, and Rutström (2007). Note:

all currency units are in EURO at the time of the experiment 1 USD = 0.86 EURO. Note that the returns

have been scaled up by a thousand tokens. a) Not shown to participants; b) Assuming a power utility

function U(x) = (1-r)-1 x1-r.

To analyze the data obtained in terms of coefficients of risk aversion we

assume under EUT the subjects’ utility function to have the following CRRA form:


50

U(x) = (1-r)-1 x1-r, where x is the lottery price (investment return) and r ≠ 1 the parameter

of risk aversion to be estimated. With this functional form, r =0 denotes risk-neutral

behavior, r >0 denotes risk aversion, and r <0 denotes risk-loving behavior. By

minimizing the difference in expected utilities obtained from option A and option B we

can calibrate the open CRRA interval in the last column of Table 2.17. In the iMPL the

subjects are presented a second table with probabilities altering in-between the

switching point of the first basic MPL. Hence, all participants are presented a

maximum of twenty choices i.e. two tables with ten rows each. Assume, for example,

that a subject switches in the first table in the third row from A to B (note that this is

the same as to say the subject has chosen two safe choices). This choices result in a

risk aversion coefficient within the interval between -0.96 and -0.49 in the first table,

i.e. is risk loving. The second table that is shown to the participant would then consist

of ten rows and two choices A and B with increasing probabilities ranging from 21%

to 30% for option A and 79% to 70% for option B, respectively. When adding the

second table and assuming the switching point is here also at row three, the risk

aversion coefficient interval is narrowed down and is now located within the interval

between -0.84 and -0.798. To analyze the subject’s degree of risk aversion we follow

Harrison, Lau, and Rutström (2007) and calculate the mid-point of the CRRA interval.

Consequently the iMPL, compared to the standard MPL, allows a richer

characterization of the utility function and thus a more refined elicitation of the true

risk attitude (Harrison, Lau, and Rutström, 2007).

7 Note that these CRRA intervals are the same as reported by Holt and Laury (2002) (Table 3).

8 The full table of resulting CRRA intervals is available upon request from the authors.


51

In order to estimate the individuals’ parameter of risk aversion we are

essentially estimating the likelihood to switch from option A to option B in each row.

The stochastic choice process specifies the likelihood of choosing one option given an

alternative option (for refinements on stochastic choice processes see Harrison and

Rutström, 2008). Under EUT the expected utility of each outcome k in each lottery i is

the probability weighted utility of each outcome in each lottery: EUi=∑k=1,K (pk×Uk),

with pk being the probabilities for each outcome. Following Abdellaoui, Driouchi, and

L’Haridon (2011); Andersen et al. (2008) and Holt and Laury (2002) we use the Luce

error9 specification to estimate the likelihoods conditional on the model:

∇EU=EUB(1/µ)

/(EUA(1/μ)

+EUB(1/μ)

), where EUA is the expected utility for ‘Option A’,

EUB is the expected utility for ‘Option B’ and µ is a structural noise parameter. This

enables us to include individuals’ choices that are not consistent with standard EUT

assumptions when estimating r. The log-likelihood of the risk aversion response,

conditional on the EUT and CRRA specifications being true, depends on the estimates

of r and µ. The likelihood of risk aversion can be written as

lnL(r,μ;y,X)=∑i((ln(∇EU|yi=1)+(ln(1-∇EU|yi=-1)), where yi = 1 (−1) denotes the

choice of the option B (A) in risk aversion task i and X is a vector of individual

characteristics.

Following Harrison, Lau, and Rutström (2007) the iMPL uses the same

incentive logic as the MPL. The participants were asked prior to the iMPL to answer a

control question about the payoff procedure, which ensures that all participants

9 See Harrison and Rutström (2008) for refinements on stochastic choice processes as well as a

detailed comparison of Fechner and Luce error specification.


52

understood the payoff structure and in turn incentive compatibility of the iMPL. For

10% of all participants one row is chosen randomly from the first table to be relevant

for payoff. Depending on the subjects choices and a randomly chosen number between

1 and 100 (reflecting the probabilities of payoff for option A and B) the individual

payoff is determined. If the row chosen at random is not the row where the subject

switched from A to B the payoff determining process ends here (identical to the MPL

payoff procedure). If the row chosen is the row that the subject switched at, another

random draw is made to pick a row in the second table that the subject was presented

with. The subject’s choice in the second drawn row is then relevant for payoff and the

procedure to determine the payoff is the same as described above10

. At the end of the

experiment, for 10% of the subjects, one choice was randomly selected to be played

out for real11

. According to Baltussen et al. (2012) this between-subjects random

incentive system, reduces the probability of real payoff for every task, possibly

inducing lower task motivation. However, we opted for this incentive system because

it allows higher prizes to be awarded to the subjects selected, which may improve

motivation and reduces the high administrative costs related to paying each participant

10

Participating students were asked to leave their unique university e-mail address enabling the

researcher to contact them for the payoff. The random draws of winners and corresponding

rows where carried out immediately after the participation deadline (average two weeks

period). The winners were invited to pick up their prices in the faculty library, in cases when

this was not possible payments where contacted and asked for further details to arrange a

transfer of the prize to the student’s bank account. 11

The expected return for each participant is 4 € for approximately 20 minutes time spent on

the survey. The average hourly wage rate of students employed as assistants at University

amounts to 8.50 €, which is used as a reference for opportunity cost of participation.

Consequently, the expected return of participating exceeds the opportunity cost, leading to an

incentive compatible iMPL.

2.4 Inconsistencies and contextualization 53

53

in a large online survey12

. Additionally, this payoff structure is adopted by several

researcher in the field like Maart-Noelck and Musshoff (2013) and Vollmer, Hermann,

and Musshoff (2017).

2.4 Inconsistencies and contextualization

Different ways of inconsistent behavior, that is behavior not consistent with

assumptions made in EUT, within the above presented iMPL are possible: i)

inconsistent response behavior is revealed if more than one switching point between

option A and B is observed; ii) inconsistent behavior is indicated by “backwards”

choices, i.e. switching in the other direction from option B in the first row to option A

in the following rows (Lévy-Garboua et al., 2012, Holt and Laury, 2002) and iii) as the

last set of choices is commonly a control question with option B clearly dominating

option A, a subject choosing A in all 10 rows is also thought of behaving inconsistent.

Because in the last row option B results with certainty in a higher payoff than option A

(see also Table 2.1 for an example). Note that in the iMPL there is a possibility of

inconsistent behavior either in the first or in the second table. Both cases are in the

following treated as within-method inconsistencies.

Based on the findings of Lévy-Garboua et al. (2012) and Holt and Laury

(2002) we avoid excessively high inconsistency rates by showing probabilities

simultaneously (i.e. the full table of choices at once) and using high payoff. The payoff

are identical to the high payoff treatment of Holt and Laury (2002), this is the original

12

See Baltussen et al. (2012) for an in depth discussion on the application of different incentive

systems.


54

lottery payoff X 20. Note that in order to construct a realistic agricultural decision the

returns in both frames (general lottery and agricultural decision) have been scaled up

by thousand. The participants were informed about the exchange value for the real

payoff in the control question prior to the iMPL.

We use two different contextual settings of the iMPL to analyze the effects of

involvement. They are as follows: First, the traditional wording according to Holt and

Laury (2002) of a gambling choice between two lotteries A and B with different payoff

and associated probabilities. In the further this is referred to as ‘general lottery’

frame13

. Second an agricultural decision with investment options A and B with

different returns and associated probabilities. This is in the further referred to as the

‘agricultural decision’ frame. Note that no time components have been included in the

task or task description. In contrast, it was very clear to the participants that payoffs are

made shortly after the experiment was conducted (for both tasks). Two pre-test

sessions with 19 students did neither reveal difficulties with respect to the experiments

payoff structure or framing, nor indicated a misunderstanding regarding the time

dimension of the agricultural investment decision. Thus, there is no evidence for time

related biases in the agricultural investment frame (Deck, Lee, and Reyes, 2014).

Nevertheless, investment decisions are undoubtedly closer to the real decisions

subjects face in there every day life, justifying the chosen comparison. The specific

application to agriculture makes use of the educational background of the participants.

13

The exact wording of the general lottery task is as following: “In each row of the following

table you can choose between two lotteries (A and B). With certain chances/ probabilities you

get for lottery A a payoff of 40.000 € or 32.000 € and for lottery B a payoff of 77.000 € or

2.000 €. Please decide between lottery A and B for every row of the table.”


55

The wording of the agriculture specific question reads as following: “Assume that after

successful completion of your studies you are offered to make an agricultural

investment. Here you will get with different associated probabilities for investment A a

return of 40,000 € or 32,000 € and for investment B a return of 77,000 € or 2,000 €.

You can choose in the following table in each row between the two investment options

(A or B).”

In order to analyze the effect of task involvement on within-method

consistency we compare the frequencies of inconsistent answers in the different iMPL

frames depending on the individual task involvement. Additionally, we compare the

structural noise µ when estimating the risk preference parameter r for both frames and

differences in context involvement using standard maximum likelihood procedures

(following Harrison and Rutström, 2008).

To analyze whether between-method consistency increases with increasing

context involvement we compare the correlation of risk aversion coefficients compiled

using the two elicitation methods described above (iMPL and self-assessment14

). Using

the methodological steps suggested by Olkin and Finn (1995) and Steiger (1980), we

compare the correlation of both risk elicitation methods in the two iMPL frames. The

same approach is used to test if task involvement influences between-method

consistency.

14

To account for the ordinal structure of responses from Likert scale questions, we use rank

correlations throughput the entire paper.

2.5 Sample description and results 56

56

2.5 Sample description and results

We obtained 370 answers and 156 complete questionnaires from Bonn University and

194 answers with 96 complete questionnaires from the South Westphalian University

of Applied Sciences leading to a total of 252 complete questionnaires (response rate of

34% and 15% complete responses). After the data cleansing process 244 surveys

remained15

. Due to strict data protection policies in both universities, only information

on the agricultural students’ gender was available. At Bonn University 43% male and

57% female agricultural students are enrolled, this is reflected in our sample. The

consistent answers amount to 195. Among them, 95 were randomly assigned to the

general lottery and 100 students were randomly assigned to the agricultural decision.

Table 2.2 shows the summary statistics for the total sample, and Table 2.3 shows the

summary statistics by the general lottery and the agricultural decision sample

separately.

The mean CRRA interval mid-point indicates an average risk aversion

coefficient of 0.57, which is in line with other researchers’ findings (see Harrison, Lau,

and Rutström, 2007). Furthermore, the self-assessment of the participants risk attitude

is with an average of 4.64 also pointing towards risk aversion. Context involvement is

defined as the personal relevance of the task for the decision maker. We calculate a

context involvement score based on the student’s agricultural involvement (referring to

the specific agricultural contextualization used). This score includes the following

15

Participants not enrolled in agricultural studies and non-German students were excluded to

eliminate biases due to different educational and cultural differences we are not accounting

for.


57

factors: rural origin, farm upbringing, parents are farmers, succession of farm holding

intended, agricultural internship, vocational training, and obtained agricultural

education certificate. The average context involvement score is 1.74 points. The

average task involvement measured by the time spent on the iMPL is 3.40 minutes, and

the time spent on the self-assessment of risk preferences 0.43 minutes (see Table 2.2

for summary statistics of variables, a detailed description of variables can be found in

Appendix A.2 Table 2.A.1). When comparing the sample with the general lottery

framing and contextualized lottery we see lower average risk aversion in the general

lottery framed iMPL. Context involvement scores are on average higher in the sample

randomly assigned to the contextualized iMPL16

(see Table 2.3).

Table 2.2: Summary statistics by within-method consistency

Total sample

Within-method

consistent

Within-method

inconsistent

N=244 N=195 N=49

mean sd NA mean sd NA mean sd NA

Self-assessment 4.65 2.66

4.54 2.61 5.08 2.83

CRRA interval mid-

point 0.53 0.70 5 0.61 0.55 0.15 1.09 5

Risk literacy 2.85 1.16

2.94 1.14 2.49 1.21

Gender (female) 0.49 0.50

0.50 0.50 0.47 0.50

Optimism 0.69 1.24

0.72 1.20 0.55 1.37

Age 24.80 2.54

24.82 2.55 24.71 2.53

Education mother 4.56 1.88 5 4.69 1.90 4 4.04 1.69 1

Context involvement

score 1.74 1.83

1.65 1.81 2.08 1.87

Time iMPL 3.40 2.63

3.53 2.32 2.88 3.60

Time self-assessment 0.43 0.62 0.50 0.77 0.29 0.16

16

We have further estimated a binary logit model using demographic characteristics of

participants as independent variables and inconsistent behavior as dependent variable.

However, this analysis did not result in statistically significant coefficient estimates (results

are available upon request).


58

Table 2.3: Summary statistics by frame and within method consistency

General lottery frame Agricultural decision frame

Within-method

consistent

Within-method

inconsistent

Within-method

consistent

Within-method

inconsistent

N=127 N=95 N=32 N=117 N=100 N=17

mean sd NA mean sd NA mean sd NA mean sd NA mean sd NA mean sd NA

Self-assessment 4.64 2.70 4.59 2.72 4.78 2.70 4.66 2.63 4.49 2.53 5.65 3.06

CRRA interval mid-

point 0.48 0.79 2 0.63 0.51 0.00 1.23 3 0.57 0.58 3 0.59 0.58 0.46 0.62 3

Risk literacy 2.89 1.18 3.04 1.16 2.44 1.13 2.80 1.15 2.84 1.12 2.59 1.37

Gender (female) 0.51 0.50 0.53 0.50 0.47 0.51 0.47 0.50 0.47 0.50 0.47 0.51

Optimism 0.68 1.23 0.72 1.23 0.56 1.22 0.70 1.25 0.73 1.18 0.53 1.66

Age 24.69 2.58 24.73 2.52 24.56 2.78 24.91 2.51 24.90 2.59 25.00 2.03

Education mother 4.41 1.86 3 4.62 1.91 2 3.77 1.54 4.71 1.89 2 4.74 1.90 2 4.53 1.87

Context involvement

score 1.58 1.67 1.52 1.61 1.77 1.85 1.90 1.98 1.77 1.98 2.68 1.79

Time iMPL 3.04 1.98 3.28 2.12 2.33 1.29 3.79 3.15 3.77 2.49 3.91 5.82

Time self-assessment 0.45 0.66 0.49 0.73 0.31 0.15 0.49 0.77 0.51 0.81 0.27 0.18


59

In order to analyze the effect of task and context involvement on within-

method consistency we first compare the frequencies of inconsistent answers in the

different iMPL frames (see Table 2.4). A total of 49 students (20%) answered

inconsistently. This is comparable to inconsistency rates found among student samples

by Lévy-Garboua et al. (2012) (36%) and Holt and Laury (2002) (13%). We group the

students into a “high task involvement” group and a “low task involvement” group

according to the time they needed to complete the iMPL. The cut-off value is the

median time needed by the whole sample to complete the task. We see clearly more

cases of inconsistent behavior in the group of students with low task involvement

(28%). In the high task involvement group only 12% behave inconsistently. The null

hypothesis of independence of consistence of answers and task involvement can be

rejected at the 1% level of significance.

Table 2.4: Contingency table of inconsistencies by task involvement

Consistent Inconsistent Row total 2

nd table

inconsistent

High task involvement 107 15 122 6

Chi-square

contribution 0.928 3.684

% of total row 88% 12% 40%

Low task involvement 88 34 122 15

Chi-square


% of total row 72% 28% 48%

Column total 195 49 244

Pearson's Chi-squared test with Yates' continuity correction = 8.2738

p-value = 0.004

We did not measure the time expended on the first and second table of the

iMPL separately. Thus, it is important to note that 48% of the inconsistent subjects in


60

the low involvement group and 40% of the inconsistent subjects in the high

involvement group behaved inconsistently in the second table. Consequently, the bias

in time measurement caused by the two.

In addition, we analyze the influence of context involvement on within-method

consistency (see Table 2.5). We see more cases of inconsistent answers in the general

lottery frame (25%) than in the agricultural decision frame (15%). This result is

slightly significant (i.e. at the 5.5% level)17

, revealing a pattern of different

inconsistency rates related to the different iMPL frames. Our approach differs from

earlier work, e.g. by Deck, Lee, and Reyes (2014), by placing the agricultural decision

task in a subject specific setting. Based on this further specification of the framing, we

can show that within-method consistency increases with context involvement.

Table 2.5: Contingency table of inconsistencies by context involvement

Consistent Inconsistent Row total

General lottery 95 32 127

Chi-square


% of total row 75% 25% 52%

Agricultural decision 100 17 117

Chi-square


% of total row 86% 15% 48%

Column total 195 49 244

Pearson's Chi-squared test with Yates' continuity correction = 3.6784

p-value = 0.055

Furthermore, we analyze maximum likelihood estimates of the individual risk

aversion coefficient r and the Luce noise parameter µ from our experiments. In this

17 The same pattern is observed if analyzing the samples from both universities individually. Results at Bonn University show a highly significant difference (at the 5% level), while the results from the South Westphalia University of Applied Sciences do not.


61

analysis, we also include the participants’ agricultural context involvement (see Table

2.6). Our estimates in both frames show that there is significant noise within both

frames. However, there is a larger estimated noise for the general lottery task than the

agricultural decision task. Additionally, there are larger estimates of noise for students

with a lower than average context involvement. This is consistent with our prior that

the general lottery task relates less to the subjects’ real world decision making process

and in turn triggers less analytic/systematic thinking.

Table 2.6: Estimates of risk aversion and Luce structural noise parameter in different iMPL

frames

Estimate Standard

error

95% confidence

interval

General lottery

High context involvement

r 0.595 0.123 0.354 - 0.835

µ 0.048 0.022 0.004 - 0.091

Low context involvement

r 0.668 0.274 0.130 - 1.205

µ 0.004 0.127 -0.244 - 0.252

Agricultural decision

High context involvement

r 0.407 0.098 0.215 - 0.598

µ 0.055 0.011 0.033 - 0.077

Low context involvement

r 0.661 0.066 0.531 - 0.790

µ 0.019 0.035 -0.050 - 0.088

Note that for this estimation only the first switching point in the first table considered. r reflects the

estimated risk preference parameter and µ the estimated structural noise.

Consequently our findings support the hypothesis that within-method

consistency increases with increasing task and context involvement.


62

When analyzing between-method inconsistencies first we examine the

correlations of both risk elicitation methods in the two frames (note that we include

only within-method consistently behaving subjects in this analysis). Table 2.7 shows

that for both frames of the iMPL the correlation coefficient with the result of the self-

assessment task is negative. Thus, herein both risk elicitation tasks point towards the

same risk preference direction. Recall that resulting from the structure of both

elicitation methods a lower value in the self-assessment and a higher value in the iMPL

indicate risk aversion. However, the correlation of the risk preferences derived in the

general lottery sample and the self-assessment task (-0.090) and the correlation of the

risk preferences derived in the agricultural decision sample and the self-assessment

task (-0.028) do not significantly differ from each other. Therefore, we reject the

hypothesis that both correlations obtained from independent samples are different.

Table 2.7: Spearman's rank correlations between CRRA interval mid-points in different iMPL

frames and self-assessment

rs

General lottery -0.090

Agricultural decision -0.028

Fishers' z-value 0.43

p-value 0.64

To test if between-method consistency is driven by the effort spent on the risk

elicitation tasks we include the subjects’ task involvement in our analysis of between-

method inconsistencies. As described above, we split our sample in two independent

groups with low and high task involvement. Here we distinguish the two groups by the

median of the total time spent on both risk elicitation methods. The correlation of the

risk aversion coefficients of the group with low task involvement is positive but not


63

significant (+0.037). Thus, here both risk preference parameters are not consistently

measuring risk averse or risk loving preferences, i.e. there are more between-method

inconsistencies in this group. In the group with high task involvement we find contrary

results. The risk aversion coefficients are here negative correlated (-0.182) at a 10%

significance level, i.e. both risk preference parameter point towards the same direction

in this group. Between-method inconsistency is here significantly lower. We cannot

reject the hypothesis that both correlation coefficients obtained from independent

samples are different. We find a statistically significant difference of the correlation of

the CRRA mid-point and self-assessment for students with higher task involvement

(see left column in Table 2.8).

Table 2.8: Spearman's rank correlations between CRRA interval mid-points in different iMPL

frames and self-assessment for different levels of task involvement

Total sample

General

lottery

Agricultural

decision

High task involvement -0.182* -0.321** -0.064

Low task involvement 0.037 0.106 -0.002

Fishers' z-value 1.59 2.07 0.30

p-value 0.11 0.04 0.77

Note: ∗ and ∗∗ denote 10% and 5% significance levels, respectively

Furthermore, we analyze if these differences in between-method

inconsistencies, depending on the individuals’ task involvement, are more severe in the

general lottery frame than in the agricultural decision frame. We find that in both

frames the correlation coefficients of iMPL and self-assessment are negative for

subjects with high task involvement. For subjects with low task involvement in the

general lottery frame we find a positive correlation of the risk preference coefficients

from both risk preference elicitation methods, thus although not significant there is

2.6 Conclusion

64

evidence for between method inconsistency. Contrarily the correlation of the two risk

preference coefficients is negative in the agricultural decision framed task, indicating

less between method inconsistencies (Table 2.8). Nevertheless, we find that this

difference in the two frames diminishes in the high task involvement group. When

looking at the significance of the difference of the two independent groups (low and

high task involvement) we find that it is only significant in the general lottery frame

(Fishers’z value = 2.07, p-value=0.04). Between-method inconsistencies are not

significantly depending on task involvement in the agricultural decision setting. Thus,

if context involvement is triggered, task involvement does not influence between-

method inconsistencies. Framing effects play only a minor role when analyzing

between-method consistency.

2.6 Conclusion

The subjective risk attitude is decisive for a wide range of decisions taken by economic

actors. Over the last decades, a wide range of approaches has been introduced to elicit

subjective risk attitudes. However, these elicitation approaches are plagued by

inconsistent responses by participants. High inconsistency rates can provoke biased

risk preference interpretation and an unavoidable reduction of the explanatory power of

the analysis. Thus, the reduction of inconsistent behavior is crucial to improve our

understanding of risk preferences. We provide the first study that investigates the role

of the decision maker’s involvement and how this relates to contextualization of

experiments. To this end, we analyze data gathered in experiments considering the

2.6 Conclusion

65

subjects’ specific differences in context involvement. The methods applied comprise a

self-assessment of risk preferences and an incentive-compatible iMPL.

We find evidence that subjects’ context and task involvement influence

inconsistencies. More specifically, we find that within-method inconsistencies are

reduced with increasing task and context involvement. In addition, between-method

inconsistencies decrease with increasing task involvement. Adding the subjects’

specific contextualization of the risk elicitation method to the analysis of between-

method consistency, we find that the importance of task involvement to increase

between-method consistency diminishes with increasing contextual involvement.

Furthermore, we find that subject specific contextualized elicitation improves

accuracy. Thus, we conclude that the contextual embeddedness of a decision making

problem is crucial when trying to analyze risk preferences of a specific subject group.

Our study was restricted to the analysis of the influence of context and task

involvement on between-method inconsistencies and on within-method inconsistencies

at one point in time. Further research should also include a time dimension to test for

the influence of task and context involvement on within-method inconsistency across

time. Furthermore, the analysis of between- method inconsistencies can be extended by

increasing the number of compared risk preference elicitation methods. Besides,

changing the task wording other exogenous variations like the stake sizes, changes in

the visualization of the task or changes of the prominence of the specific task in the

experimental design as a whole could be used to extend findings on ways to trigger

involvement. In addition to comparing differences in inconsistencies resulting from

2.6 Conclusion

66

adaptations of the standard MPL further research should concentrate on comparing

noise estimates for different underlying theoretical concepts. For example, noise

parameters estimated based on the rank dependent utility model could be compared

with the estimated noise parameter assuming EUT.

Although students are the most convenient subjects for researchers in

experimental economics, there have been difficulties when trying to project students

risk preferences to real decision makers (see Carpenter, Burks, and Verhoogen, 2005).

We have shown that with increasing involvement students perform better in risk

elicitation tasks in terms of consistency. Future research should concentrate on finding

ways to trigger students’ task and context involvement if they are used as a

convenience group in experimental economics. Additionally, research should also

concentrate on showing that risk preferences of students with higher context

involvement could project risk preferences of real decision makers. In turn, this can

lead to better predictions of real world decisions and thus improve policy analysis.

Finally, we have shown that by using a real life and subjective context related

MPL, involvement can be triggered and consequently the problem of inconsistencies

and misinterpretations caused by it can be reduced.

2.7 References

67

2.7 References

Abdellaoui, Mohammed, Ahmed Driouchi, and Olivier L’Haridon. 2011. "Risk

aversion elicitation: reconciling tractability and bias minimization." Theory and

Decision 71 (1):63-80.

Alatas, Vivi, Lisa Cameron, Ananish Chaudhuri, Nisvan Erkal, and Lata Gangadharan.

2009. "Subject pool effects in a corruption experiment: A comparison of Indonesian

public servants and Indonesian students." Experimental Economics 12 (1):113-132.

Alekseev, Aleksandr, Gary Charness, and Uri Gneezy. 2017. "Experimental methods:

When and why contextual instructions are important." Journal of Economic

Behavior & Organization 134:48-59.

Andersen, Steffen, Glenn W Harrison, Morten Igel Lau, and E Elisabet Rutström.

2006. "Elicitation using multiple price list formats." Experimental Economics 9

(4):383-405.

Andersen, Steffen, Glenn W. Harrison, Morten I. Lau, and Elisabet E. Rutström. 2008.

"Eliciting Risk and Time Preferences." Econometrica: Journal of the Econometric

Society 76 (3):583-618.




Baltussen, Guido, Thierry G. Post, Martijn J. van den Assem, and Peter P. Wakker.

2012. "Random incentive systems in a dynamic choice experiment." Experimental

Economics 15 (3):418–443.

Bruner, David M. 2009. "Changing the probability versus changing the reward."

Experimental Economics 12 (4):367-385.

Carbone, Enrica, and John D. Hey. 2000. "Which error story is best?" Journal of Risk

and Uncertainty 20 (2):161-176.

Carpenter, Jeffrey, Stephen Burks, and Eric Verhoogen. 2005. "Comparing students to

workers: The effects of social framing on behavior in distribution games."

Research in Experimental Economics 10:261-290.



2.7 References

68

Charness, Gary, and Angelino Viceisza. 2016. "Three Risk-elicitation Methods in the

Field: Evidence from Rural Senegal." Review of Behavioral Economics 3 (2):145-

171.

Cokely, Edward T., Mirta Galesic, Eric Schulz, Saima Ghazal, and Rocio Garcia-

Retamero. 2012. "Measuring risk literacy: The Berlin numeracy test." Judgment

and Decision Making 7 (1):25.

Coppola, Michela. 2014. "Eliciting risk-preferences in socio-economic surveys: How

do different measures perform?" The Journal of Socio-Economics 48:1-10.

Crosetto, Paolo, and Antonio Filippin. 2015. "A theoretical and experimental appraisal

of four risk elicitation methods." Experimental Economics 19 (3):1–29.

Csermely, Tamás, and Alexander Rabas. 2017. "How to reveal people’s preferences:

Comparing time consistency and predictive power of multiple price list risk

elicitation methods." Journal of Risk and Uncertainty 53 (2):107-136.

Dave, Chetan, Catherine C. Eckel, Cathleen A. Johnson, and Christian Rojas. 2010.

"Eliciting risk preferences: When is simple better?" Journal of Risk and


Deck, Cary, Jungmin Lee, and Javier Reyes. 2014. "Investing versus gambling:

experimental evidence of multi-domain risk attitudes." Applied Economics Letters

21 (1):19-23.




(3):522–550.

Eckel, Catherine C., and Philip J. Grossman. 2002. "Sex differences and statistical

stereotyping in attitudes toward financial risk." Evolution and Human Behavior 23

(4):281–295.

Harrison, Glenn W., Morten I. Lau, and Elisabet E. Rutström. 2007. "Estimating risk

attitudes in denmark: A field experiment." The Scandinavian Journal of Economics

109 (2):341-368.

Harrison, Glenn W., and Elisabet E. Rutström. 2008. "Risk aversion in the laboratory."

In Risk aversion in experiments, edited by James C. Cox and Glenn W. Harrison,

41-196. online: 08 Mar 2015: Emerald Group Publishing Limited.

2.7 References

69

Hirschauer, Norbert, Oliver Musshoff, Syster C. Maart-Noelck, and Sven Gruener.

2014. "Eliciting risk attitudes – how to avoid mean and variance bias in Holt-and-

Laury lotteries." Applied Economics Letters 21 (1):35–38.



Holt, Charles A., and Susan K. Laury. 2005. "Risk aversion and incentive effects: New

data without order effects." American Economic Review:902–904.

Jacobson, Sarah, and Ragan Petrie. 2009. "Learning from mistakes: What do

inconsistent choices over risk tell us?" Journal of Risk and Uncertainty 38 (2):143–

158.

Kahneman, Daniel. 2003. "Maps of bounded rationality: Psychology for behavioral

economics." American Economic Review 93 (5):1449-1475.



Levin, Irwin P., Sandra L. Schneider, and Gary J. Gaeth. 1998. "All frames are not

created equal: A typology and critical analysis of framing effects." Organizational

Behavior and Human Decision Processes 76 (2):149-188.

Lévy-Garboua, Louis, Hela Maafi, David Masclet, and Antoine Terracol. 2012. "Risk

aversion and framing effects." Experimental Economics 15 (1):128-144.

Maart-Noelck, Syster C., and Oliver Musshoff. 2013. "Measuring the risk attitude of

decision-makers: are there differences between groups of methods and persons?"

Australian Journal of Agricultural and Resource Economics 58 (3):336–352.

McElroy, Todd, and John J. Seta. 2003. "Framing effects: An analytic–holistic

perspective." Journal of Experimental Social Psychology 39 (6):610-617.

Meraner, Manuela, Robert Finger, and Oliver Musshoff. submitted. "Using

involvement to reduce inconsistencies in risk preference elicitation."

Stanovich, Keith E., and Richard F. West. 1999. "Discrepancies between normative

and descriptive models of decision making and the understanding/acceptance

principle." Cognitive Psychology 38 (3):349-385.

Starmer, Chris. 2000. "Developments in non-expected utility theory: The hunt for a

descriptive theory of choice under risk." Journal of Economic Literature 38

(2):332-382.

2.7 References

70

Steiger, James H. 1980. "Tests for comparing elements of a correlation matrix."

Psychological Bulletin 87 (2):245-251.

Thoma, Carmen. 2015. "Under- versus overconfidence: an experiment on how others

perceive a biased self-assessment." Experimental Economics 19 (1):218-239.

Tversky, Amos, and Daniel Kahneman. 1986. "Rational choice and the framing of

decisions." The Journal of Business 59 (4):251–278.

Vollmer, Elisabeth, Daniel Hermann, and Oliver Musshoff. 2017. "Is the risk attitude

measured with the Holt and Laury task reflected in farmers’ production risk?"

European Review of Agricultural Economics:1-26.

Wagner, Gert G., Joachim R. Frick, and Jürgen Schupp. 2007. "The German Socio-

Economic Panel Study (SOEP): Scope, Evolution and Enhancements." Schmollers

Jahrbuch 127 (1):139–169.

Wilcox, Nathaniel T. 2008. "Stochastic models for binary discrete choice under risk: A

critical primer and econometric comparison." In Risk aversion in experiments,

edited by James C. Cox and Glenn W. Harrison, 197-292. online: 08 Mar 2015:

Emerald Group Publishing Limited.

2.8 Appendix 2.A

71

2.8 Appendix 2.A

Table 2.A.1: Variable description

Risk preference

elicitation method Measurement description

CRRA interval mid-

point iMPL CRRA interval mid-point

Self-assessment 0 if very risk averse; …; 10 if very risk loving

Variable Name Variable description

Gender 1 if female

Age Years

Optimism Difference of life satisfaction in a year and life satisfaction today

(both measured on a scale from 0 to 10)

Risk literacy score 1 = poor numerical reasoning; 2 = rather poor numerical reasoning;

3 = good numerical reasoning; 4 = very good numerical reasoning

(according to Cokely et al., 2012)

Education mother Mothers highest education according to the German schooling

system: 1if no degree obtained;…; 9 if PhD degree obtained

Involvement score Sum of involvement factors described below

Rural origin 0.5 if area of growing up has less than 20,000 inhabitants

Growing up on farm

holding

1 if grew up on a farm

Parents are farmers 1 if parents are farmers

Succession of farm

holding intended

0.5 if probably no succession is intended; 1 if probably succession

is intended; 2 if succession is intended

Agricultural

internship

0.5 if internship time is less or equal to 6 months; 1 if internship

time is more than 6 months

Vocational training 1 if agriculture specific vocational training obtained

Agricultural school 1 if three year agricultural school degree

Master exam 1 if five year agricultural school degree (master)

Higher agricultural

education

1 if higher agricultural education obtained

Time iMPL Time spent on iMPL in minutes

Time Self-assessment Time spent on self-assessment of risk preferences in minutes

2.9 Appendix 2.B

72

Figure 2.A.1: Distribution of the CRRA interval mid-point in two different iMPL frames

Figure 2.A.2: Distribution of self-assessment task

2.9 Appendix 2.B

First screen

Welcome to the experiment!

In the following we will offer you various situations and options to choose from. We

would like to get to know something about your behavior in different

situations/scenarios. There are no 'right' or 'wrong' decisions!

2.9 Appendix 2.B

73

For all participants there is a chance to be drawn at random for a win of 87 €. We will

inform you about your payoff via e-mail. The payoff of the win will be carried out

immediately after the evaluation of the experiment.

The experiment will take approx. 20 minutes. Of course, your data will be treated

confidentially and the data will be evaluated anonymously. For further inquiries please

contact: [email protected].

Second screen

Please read carefully through the following description of your chances to win:

What can you win? The payoff for each participant can amount up to 87 € (first part:

77 € + second part: 10 €).

How can you win? In the first part of the experiment you will be presented a table

with ten rows. For each row you will have to choose between option A and option B.

The decision (row), that will be relevant for your payoffs, will be determined by the

first draw out of a lottery with ten balls. In the case that row 4 will be identified, you

were asked to choose between option A (40% probability/chance 40.000 € und 60%

probability/chance 32.000 €) and option B (40% probability/chance 77.000 € and 60%

probability/chance 2.000 €). Your win will be multiplied by the factor 1/1.000.

If we assume that you have chosen option B in the randomly selected row 4. In the

second draw, the numbers 1 to 4 (= 40% chance) lead to a payoff of 77 €, the numbers

5-10 (= 60% chance) result in a payoff of 2 €. In the second part we will ask you to

solve some arithmetic questions. For the correct answer to these tasks, the participants,

who were selected as winners, receive additionally 10 €.

Who can win? 10% of all participants will be drawn at random to receive the

payment.

Third screen: control question

To make sure, that you understood the method of payment for the reward of your

participation, please answer the following question:

Please assume that you were drawn randomly as one of the winners. In the first draw,

which serves to identify the row, that will be relevant for the payment, 4 out of 10 was

drawn. This means that the decision row 4 will be relevant for your payment. Assume

furthermore, that you have chosen option A in the relevant decision (marked with the

blue dot in the table on the right).

The second draw results in number 7. What is the amount of your payoff?

(1.000 € in the lottery = 1 € payoff).

(Right answer not shown to participants: 32)

2.9 Appendix 2.B

74

Figure 2.B.1: Third screen: control question

Only one of the following two screens is shown:

Fourth screen A: General lottery task

In each row of the following table you can choose between two lotteries (A and B).

With certain chances/ probabilities you get for lottery A a payoff of 40.000 € or

32.000 € and for lottery B a payoff of 77.000 € or 2.000 €. Please decide between

lottery A and B for every row of the table.

Fourth screen B: Agricultural decision task

Assume that after successfully completing your studies you are offered to make an

agricultural investment. Here you will get with different associated probabilities for

investment A a return of 40,000 € or 32,000 € and for investment B a return of

77,000 € or 2,000 €. You can choose in the following table in each row between the

two investment options (A or B). Please decide between investment A and B for every

row of the table.

2.9 Appendix 2.B

75

Figure 2.B.2: Fourth screen: General lottery task

Fifth screen: Self-assessment

How do you see yourself: are you generally a person who is fully prepared to take risks

or do you try to avoid taking risks? Please tick a box on the scale, where the value 0

means: `not at all willing to take risks' and the value 10 means: `very willing to take

risks'.

Figure 2.B.3: Fifth screen: Self-assessment task

76

Chapter 3

Risk perceptions, preferences and management strategies:

Evidence from a case study using German livestock

farmers*

* Meraner, M. and R. Finger (accepted for publication at the Journal of Risk Research)

77

Abstract

We analyze factors affecting farmers’ choice accounting for farm, farmer

and household characteristics as well as elicited risk perception and risk

preferences. We consider three alternative hypothetical methods for

assessing risk preferences to test the stability and behavioral validity of

them. Our case study focusses on livestock farmers in the German region

North-Rhine-Westphalia. We find that risk preferences are context

depending, i.e. differ across different fields of farm-level decision making.

Furthermore, our analysis shows that risk averse farmers are more likely to

prioritize on-farm risk management strategies over off-farm strategies.

Moreover, higher risk perception, age, subjective numeracy, farm

succession, farm size and the proportion of rented land show significant

impact on farmers’ risk behavior.

Keywords: risk preference elicitation, risk perception, risk

management

JEL classification: D81, Q12

3.1 Introduction

78

3.1 Introduction

In agricultural production farmers are confronted with a wide and increasing range of

production, market, financial and institutional risks. Consequently, the portfolio of risk

management strategies available to farmers is large and growing, but little is known

about the farmer’s decision-making process when choosing the optimal risk

management strategies. Behavior under risk typically results from the interplay of the

perceived risk level faced by decision-makers and their own preferences towards risk.

In order to understand farmers’ choice of risk management strategies both farmers’

perception of risks and farmers’ personal characteristics (including preferences towards

risk) need to be considered.

The perception of risks (i.e. the perception of the economic environment) of

the decision maker is widely recognized to influence the decision making process

(Slovic, Fischhoff, and Lichtenstein, 1982, van Raaij, 1981). Risk perception is

determined by the objective risk an individual is exposed to and the subjective

interpretation of risks. Consequentially, risk perception is the combination of the

probability of the occurrence of an uncertain event and the consequential negative

impact (Slovic, Fischhoff, and Lichtenstein, 1982).

Accounting for risk preferences is crucial to better understand farmers’

decisions. The literature on risk preference elicitation methods has grown rapidly over

the last decades (see Charness, Gneezy, and Imas, 2013 for an extensive overview).

Most prevalent methods are based on hypothetical or non-hypothetical lottery-choices

(Eckel and Grossman, 2008, Holt and Laury, 2002, Gneezy and Potters, 1997, Crosetto

3.1 Introduction

79

and Filippin, 2013) or survey questions (Weber, Blais, and Betz, 2002, Dohmen et al.,

2011, Hansson and Lagerkvist, 2012). Findings on which risk preference elicitation

methods can indeed best predict real risk behavior however remain ambiguous. For

example, Dohmen et al. (2011) find that a simple general self-assessing risk question is

better suited to explain real world behavior than the commonly used multiple price list

(MPL). Moreover, Hellerstein, Higgins, and Horowitz (2013) show for a sample of US

farmers that risk preferences elicited using a MPL do not necessarily allow to explain

risk management choices. Recently, Menapace, Colson, and Raffaelli (2015) found

that using a contextualized MPL for risk preference elicitation pertains better to the

insurance uptake of Italian farmers than the standard non contextualized MPL. Along

these lines, Meuwissen, Huirne, and Hardaker (2001) have shown that a set of four

contextualized business questions specific to farming eliciting farmers relative risk

aversion, is correlated with risk management choices.

Literature on farmers’ choice of risk management strategies is often focusing

on the adoption of single activities such as insurances (Menapace, Colson, and

Raffaelli, 2015, Finger and Lehmann, 2012). In reality, however, farmers use a large

portfolio of different risk management strategies in order to react to different risk

sources (Musser and Patrick, 2002, Meuwissen, Huirne, and Hardaker, 2001, van

Winsen et al., 2014). Thus, a larger set of risk management actions and their

interrelations needs to be considered based on a holistic portfolio approach. Studies

that address this aspect usually consider a limited set of risk management tools such as

on-farm agricultural diversification and insurance in Hellerstein, Higgins, and

3.1 Introduction

80

Horowitz (2013), insurance and hedging in Mishra and El-Osta (2002), on-farm and

off-farm diversification in McNamara and Weiss (2005) or diversification of crop

portfolios and off-farm labor allocation in de Mey et al. (2016). These studies,

however, have not explicitly accounted for elicited risk preferences. Moreover, studies

that consider a larger set of risk management strategies often apply factor analysis to

reduce the number of risk management tools included in further analysis without

theory based structuring (van Winsen et al., 2014, Meuwissen, Huirne, and Hardaker,

2001, Flaten et al., 2005). This ad-hoc categorization of risk management tools leads to

difficulties when comparing results of those studies. Hence, there is a lack of empirical

studies addressing holistic perspectives of risk management decisions and a lack of

studies investigating the power of different risk preference elicitation methods to

explain the composition of risk management choices taken at the farm-level.

In this study, we fill this research gap by investigating several risk attitude

elicitation methods and risk management strategies simultaneously. We consider three

relatively simple, easy to implement risk elicitation instruments and test their power in

pertaining to actual farmer’s decisions. More specifically, we use i) a general self-

assessment (SA) of risk preferences, ii) a contextualized version of the multiple price

list (MPL), and iii) a set of four agriculture specific business statements (BS). We use

contextualized elicitation measures in our analysis (MLP and BS) as those have been

shown to reduce within- and across- method inconsistencies in risk preference

elicitation (Reynaud and Couture, 2012).

3.2 Conceptual framework and relevant literature

81

The objective of this study is to examine the factors that influence farmers’ risk

management decisions. We consider the possibility of simultaneous utilization of

multiple risk reducing instruments and the potential correlations among those adoption

decisions. Based on this background, we i) aim to analyze how farm and farmer’s

characteristics are related to the choice of risk management strategies and ii) aim to

reveal which risk attitude parameter elicited from three different risk elicitation

methods pertains the farmers’ risk behavior best. Our empirical case study focusses on

livestock producer in the German Munster region in North-Rhine-Westphalia

(Germany). We focus on risk management strategies classified depending on the

farmers’ choice of resource shift, in three categories: on-farm agriculture, on-farm non-

agriculture, and off-farm strategies. The remainder of this paper is structured as

follows. Preceded by a literature review, the experimental design is presented.

Subsequently, the methodological approach and data are introduced. Finally, the results

are presented followed by a discussion and a concluding section.


Economic research in the field of risk behavior and risk management is often based on

expected utility theory, with the utility maximizing decision maker at its core (Meyer,

2002). This expected utility hypothesis has been criticized on descriptive grounds, i.e.

due to its lack of predictive power of real world decision-making behavior (Kahneman

and Tversky, 1979, Slovic, Fischhoff, and Lichtenstein, 1982, Savage, 1972). Thus, we

include in our assessment of the farmers’ choice of risk management strategy the


82

subjective attitude towards risk and believes regarding the probability of an uncertain

outcome occurring (i.e. risk perception). According to subjective expected utility

(Savage, 1972, Slovic, Fischhoff, and Lichtenstein, 1982), understanding the

individual’s reference frame for evaluating choices with uncertain outcomes is crucial.

In other words, the farmers’ decision-making environment together with his personal

characteristics is determining the decision-making behavior. Thus, we introduce a

framework recognizing farm, farmer and household characteristics to determine the

farmers risk management choice (left hand side in Figure 3.1). The farm holdings’

main goal is to allocate its resources, within the economic environment, in a utility

maximizing way. According to van der Ploeg and Roep (2003) the farm enterprise has

three dimensions where the limited resources (e.g. land, labor and capital) can be

allocated: on-farm agricultural, on-farm non-agricultural and off-farm. We propose to

adopt these three dimensions of farm resource allocation to the choice of risk

management strategies. In our analysis, we focus on risk management strategies related

to on-farm agriculture, on-farm non-agriculture, and off-farm decisions (see right hand

side Figure 3.1).

Figure 3.1: Factors influencing farmers’ decision-making on farm resource allocation

dimensions


83

Although no study has as yet compared determinants of adoption of those

strategies within a single portfolio, in this section we review the literature that

addresses determinants of farmers’ choices within individual categories. We find

empirical evidence that greater risk aversion is associated with a higher uptake

probability of off-farm risk management tools. Off-farm risk management strategies

can be associated with a reduction of dependency on (risky) farm income. For instance,

in a study using Swiss FADN data, de Mey et al. (2016) find that farm households that

are exposed to greater financial risks show a higher share of off-farm labor income.

Furthermore, van Winsen et al. (2014) find a positive relationship of risk aversion and

the uptake of off-farm work and off-farm investments amongst Belgian farmers.

Counter intuitively some studies find that insurance uptake is negatively correlated

with risk aversion (Menapace, Colson, and Raffaelli, 2015, Hellerstein, Higgins, and

Horowitz, 2013, Just, Calvin, and Quiggin, 1999). Some argue that the artificial nature

of most risk preference elicitation methods provokes decision-making in experimental

settings detached from real world behavior (Menapace, Colson, and Raffaelli, 2015,

Hellerstein, Higgins, and Horowitz, 2013). Harrison, List, and Towe (2007) argue that

the underlying background risk that farmers are facing leads to those already applying

insurance, as a risk management strategy are more willing to take risks in the

experiment. Furthermore, the structure of insurance schemes is argued to influence the

uptake probability more than risk aversion (Just, Calvin, and Quiggin, 1999). In other

research fields, higher risk aversion has been found to go alongside less risk reducing

behavior (e.g. Holden and Quiggin, 2016). Risk aversion may hinder or delay the


84

adoption of new technologies, as they are connected to new uncertainty and thus

provoke extra caution among more risk averse farmers. Furthermore, the uptake of

insurance for irreplaceable commodities (e.g. life at risk) has been shown to decrease

with greater risk aversion (Bommier and Le Grand, 2014). Cook and Graham (1977)

explain that rational insurance decisions aim at equalizing the marginal utility of

wealth across states of nature with irreplaceable commodities this may generate risk

taking behavior. Increasing age is expected to decrease off-farm employment

opportunities and hence to decrease the share of off-farm risk management tools in the

farmers’ portfolio (de Mey et al., 2016). Furthermore, saturating effects of age are

expected (de Mey et al., 2016), which will be considered in our empirical analysis by

including the variable in linear and squared terms. A higher level of education is

enhancing the farmers’ off-farm labor market opportunities and in turn, increases the

likelihood to work outside the farm. As a result, those farmers spend less time on self-

protection from risk and rely more on crop insurance. Additionally, farmers with

greater risk literacy (subjective numeracy is used as a proxy) are expected to have

enhanced ability to acquire and decode information which in turn is associated with

higher adoption rates of insurance contracts and participation in the commodity futures

exchange (Mishra and El-Osta, 2002, Velandia et al., 2009). Nevertheless, human

capital theory suggests that increasing education is associated with decreasing risk

aversion and hence less participation in market based risk reducing strategies (Velandia

et al., 2009, Shapiro and Brorsen, 1988). This may also hold for participation in

agricultural training acquired in workshops –a fact that will be considered in our


85

analysis. Likewise, an optimistic mindset is also assumed to be related to greater

willingness to take risks. Thus, optimistic farmers are assumed to have a smaller share

of off-farm risk management strategies in their portfolio (Dohmen et al., 2011). Risk

perception varies between individuals depending on the objective risk they are exposed

to and the subjective interpretation of risks. Risk perception is assumed to be domain

specific. Hence, we measure farmers’ risk perception in five domains. Results from

previous studies suggest that higher risk perception in all domains is associated with

higher uptake probability of risk management strategies. So far, the empirical

relationship between perceived sources of risks and responses has been ambiguous

(Flaten et al., 2005, Meuwissen, Huirne, and Hardaker, 2001, van Winsen et al., 2014).

Additionally, Menapace, Colson, and Raffaelli (2013) find that past experienced losses

influence the subjective risk perception of farmers. Thus, we assume that farmers that

experienced losses over the past five years react with multiple risk management

responses. On the farm-household level the availability of more on-farm work force as

well as farm succession and greater household size indicate the farm families’ need to

generate employment on the farm. Farm households with planned succession are

according to life cycle theory (succession effect) seeking to create a stable extra

income on the farm, consequently they are more likely to generate extra forms of

income on the farm to support the new generation of farming family (Potter and

Lobley, 1996). Additionally, there is strong evidence that farms with more available

workforce on the farm and larger household size are more likely to engage in on-farm

3.3 Experimental design

86

non-agricultural diversification strategies (Meraner et al., 2015, Benjamin and Kimhi,

2006).

Larger farm size and greater livestock are both associated with greater wealth

and thus a larger capacity to bear risks on the farm, reducing the necessity to shift

resources of the farm (Velandia et al., 2009). Larger farms in terms of size and

livestock show signs of greater specialization and are thus associated with a greater

share of on-farm agriculture related risk management strategies. Moreover, households

with a small share of rented land have higher levels of equity, which positively affects

their financial stability and in turn reduces the necessity of off-farm risk management

tools (de Mey et al., 2016, Mishra and El-Osta, 2002).


To connect farm, farmer and household characteristics with the farmers risk

management decisions, we conducted a survey amongst specialized livestock farmers

in the German region North-Rhine-Westphalia in December 2015 and January 201619

.

Farmers are a popular population subsample for conducting risk experiments, as their

profession naturally entails a large degree of risk forcing them to make regularly

decisions under risk and uncertainty (Menapace, Colson, and Raffaelli, 2015,

Herberich and List, 2012). The survey was distributed via the advisors of the local

extension service, to 256 farmers located mainly in the region of Muenster. We

included a stamped, self-addressed envelope, leading to a total of 64 responses

19

The full survey is available in the supplementary files attached.


87

(representing a 26% response rate). Focusing on a comparable group of farmers with

respect to the type of farming, geophysical characteristics of the region and available

risk management tools allows us to coherently compare strategies taken by these

farmers. We carefully designed the survey based on seven pre-tests including in-depth

interviews with young farmers and two expert feedback rounds with farm advisors,

ensuring a user-friendly layout and understandability of all questions. The paper pencil

questionnaire is structured in the following parts: i) subjective perception of risk, ii)

risk preference elicitation, iii) farmer’s characteristics, iv) household characteristics, v)

information about the farm holding, vi) risk management tools used. Farmers could

indicate to get a feedback regarding their risk perceptions, attitudes and management

strategies, as well as aggregated information on the whole sample.

In total 51 farmers (80%) of all farmers, participating requested the feedback

indicating great interest of farmers in the survey. In line with Menapace, Colson, and

Raffaelli (2013) and Reynaud and Couture (2012) this feedback report is used as a

non-monetary incentive for the participants.

Risk perception

We use an exploratory approach to measure subjective risk perception where the main

sources of risks farmers perceive to be exposed to are not clearly defined. Risk

perception can be regarded as the combination of the probability of an uncertain event

happening and the incidental impact or negative consequence (Slovic, Fischhoff, and

Lichtenstein, 1982). Consequentially, risk perception will increase when the

probability of occurrence increases, the magnitude of the impact increases or both


88

increases. To measure the farmers’ subjective risk perception we asked farmers to

score the perceived probability of 25 different risk sources on a five point scale from 1

(very unlikely) to 5 (very likely) and the perceived impact for each source on a five

point scale from 1 (very small impact) to 5 (very big impact). The perceived risk scores

are calculated by multiplying the perceived probability of occurrence with the

perceived impact (Flaten et al., 2005, Meuwissen, Huirne, and Hardaker, 2001, van

Winsen et al., 2014). The 25 risk sources included in in the survey where based on the

in-depth expert interviews with two extension service consultants and two farmers as

well as a literature study (Meuwissen, Huirne, and Hardaker, 2001, van Winsen et al.,

2014, Musser and Patrick, 2002)20

. For the subsequent analysis, we grouped them

according to five main categories of risk sources. Those categories comprise i) market

and price risks (e.g. increasing price volatility), ii) political and structural risks (e.g.

decreasing direct payments), production risks (e.g. yield volatility due to climate

change), financial risks (e.g. liquidity shortage) and other risks (e.g. shortfall of

qualified workforce) (Musser and Patrick, 2002). We calculate a risk score for each

category by taking the mean overall risk scores in each category.

Risk preferences

To elicit the farmers’ risk preferences we include the following three methods in the

survey.

Contextualized multiple price list (MPL)

20

The complete list of all 25 risk sources is included in Table 3.A.1 in Appendix 3.A.


89

The first measure of risk preferences elicited from the sample of farmers was a

contextualized MPL. The MPL is characterized by a fine gradation in the risky choices

and uses real rather than hypothetical outcomes. Following Holt and Laury (2002) we

assume constant relative risk aversion (CRRA)21

. The CRRA utility is defined as

𝑈(𝑥) = (1 − 𝑟)−1𝑥1−𝑟, where r is the CRRA coefficient. In the following analysis, we

use the CRRA interval mid-point as the farmer’s risk preference parameter (see

column 7 in Table 3.1). We create a realistic payout structure of the contextualized

MPL using payouts ranging from 5€ to 192.5€. The upscaling of payouts does not

change the original CRRA intervals used by Holt and Laury (2002) but creates an

incentive compatible MPL22

. A risk-neutral person would select option A in the first

four rows of Table 3.1 and option B in the last six rows. Nevertheless, the main

challenges with this risk preference elicitation is its relative complexity leading to

inconsistent choices and reduced predictive power (Charness, Gneezy, and Imas,

2013). Based on findings of Harrison, List, and Towe (2007), Menapace, Colson, and

Raffaelli (2015), and Hellerstein, Higgins, and Horowitz (2013) we use a modified

wording of the standard lottery to reduce complexity. More specifically, we use a

wording explicitly framed in an agricultural context to create decision-taking

approaches closer to those that farmers have previously experienced23

. To this end, we

21

The assumption of constant relative risk aversion (CRRA) has been shown to hold in the

context of medium-scale lottery in developed countries (Heinemann, 2008). 22

The expected return for each participating farmer is 9.50 €, the average time to complete the

questionnaire was estimated at 20 minutes, resulting in an hourly wage equivalent of 28.50 €. 23

The MPL question reads as following: “Assume that you are offered to make an agricultural

investment. Here you will get with different associated probabilities for investment A a return

of 100,000 € or 80,000 € and for investment B a return of 192,500 € or 5,000 €. You can

choose in the following table in each row between the two investment options (A or B).”


90

multiplied returns in the contextualized MPL by a thousand tokens. Participants were

instructed that real payouts are reconverted by the factor of one thousand. Furthermore

we reduce complexity and consequential inconsistent behavior by including a pie chart

displaying proportions next to the verbal presentation of decisions as a visual aid

(Bougherara, Gassmann, and Piet, 2011, Reynaud and Couture, 2012) (an example of

the visual presentation as well as full instructions are found in Figure 3.A.1 in

Appendix 3.A). To incentivize the MPL we follow Maart‐Noelck and Musshoff (2014)

and informed farmers that at the end of the survey period 10% of all participants are

selected for real payouts based on their choices24

.

Table 3.1: Standard payoff table MPL

Choice

Nr.

Prob. 1 vs.

Prob. 2 Option A Option B

Difference

in expected

payouts a)

Open CRRA

interval a) b)

CRRA

interval

mid-

point a)

1 10% vs. 90% 100€ vs. 80€ 192.5€ vs. 5€ 58.25 r < -1.71 -1.71

2 20% vs. 80% 100€ vs. 80€ 192.5€ vs. 5€ 41.50 -1.71 < r ≤ -0.95 -1.33

3 30% vs. 70% 100€ vs. 80€ 192.5€ vs. 5€ 24.75 -0.96 < r ≤ -0.49 -0.72

4 40% vs. 60% 100€ vs. 80€ 192.5€ vs. 5€ 8.00 -0.50 < r ≤ -0.14 -0.31

5 50% vs. 50% 100€ vs. 80€ 192.5€ vs. 5€ -8.75 -0.15 < r ≤ 0.15 0.00

6 60% vs. 40% 100€ vs. 80€ 192.5€ vs. 5€ -25.50 0.16 < r ≤ 0.41 0.28

7 70% vs. 30% 100€ vs. 80€ 192.5€ vs. 5€ -42.25 0.42 < r ≤ 0.68 0.54

8 80% vs. 20% 100€ vs. 80€ 192.5€ vs 5€ -59.00 0.69 < r ≤ 0.97 0.82

9 90% vs. 10% 100€ vs. 80€ 192.5€ vs. 5€ -75.75 0.98 < r ≤ 1.37 1.17

10 100% vs. 0% 100€ vs. 80€ 192.5€ vs. 5€ -92.50 1.37 < r

Source: Own depiction according to Holt and Laury (2002) and Harrison, Lau, and Rutström (2007). Note:

all currency units are in EURO at the time of the experiment 1 USD = 0.92 EURO. Note that the returns

have been scaled up by a thousand tokens. a) Not shown to participants; b) Assuming a power utility

function 𝑈(𝑥) = (1 − 𝑟)−1𝑥1−𝑟.

Self-assessment (SA) of risk preferences

Second, we included the following straightforward SA of the willingness to take risk:

‘On a scale from 0 to 10, where 0 means “not at all willing to take risks” and 10 means

24

This between-subjects random incentive system reduces the probability for real payout for

every task; however it allows to award higher prices to the participants and reduces the high

administrative costs related to paying each participant in a mail survey.


91

“very willing to take risks”, how would you assess your personal preference to take

risks?’ (following Dohmen et al., 2011). To ensure consistency with the results of the

other risk preference elicitation methods included in this study the self-assessment

values are inverted so that higher values imply higher risk aversion. This instrument’s

biggest advantage is its simplicity, resulting in a wide potential for collecting risk

preference measurements at a very low marginal cost (Menapace, Colson, and

Raffaelli, 2015). Dohmen et al. (2011) show that for some domains risk preferences

elicited in the SA and lottery task are highly correlated. Moreover, Maart‐Noelck and

Musshoff (2014) find significant correlations of estimated risk attitudes in both tasks

(SA and MPL) in a sample of German farmers. However, as the question is not

contextualized to a specific risk domain the applicability to predict real farm-level risk

management behavior might be limited (Menapace, Colson, and Raffaelli, 2015).

Business statements (BS)

Third, we use the following four BS related to three major sources farmers are exposed

to and to agriculture in general: “On a scale from 1 to 5, where 1 means “fully agree”

and 5 means “don't t agree” please indicate your position on the following statements:

“I am willing to take more risks than my colleagues with respect to 1. … production

risks; 2. … marketing and pricing risks; 3. … financial risks; 4. … farming in general.”

These contextualized questions follow other studies (e.g. Meuwissen, Huirne, and

Hardaker, 2001, Flaten et al., 2005, van Winsen et al., 2014, Bishu et al., 2016)

ensuring comparability of results. This method is simple, fast to complete and allows

for domain specific contextualization, i.e. directly referring to the main sources of risk


92

in agricultural practice. Since those statements measure attitude towards risks relative

to other farmers following Meuwissen, Huirne, and Hardaker (2001) we use the term

‘relative risk attitude’ in the remainder of this paper25

.

Socio economic characteristics

Our survey also included questions on farmers’ personal characteristics that comprise

the farmers’ age, level of education, participation in agricultural training sessions and

experienced past losses. Additionally, to capture the general level of optimism we

included two questions addressing the farmers’ current life satisfaction and predicted

life satisfaction in one year. We used a set of seven self-assessment questions on

numerical aptitude and preferences for numbers, adapted from Fagerlin et al. (2007) to

measure the farmers’ subjective numeracy and ability to process probabilistic

information. This subjective numeracy test is faster, avoids frustration amongst

participants and correlates strongly with risk comprehension and objective numeracy

tests (i.e. mathematical tasks) (Fagerlin et al., 2007, Zikmund-Fisher et al., 2007). On

the household level, we included information on the farms’ work force availability,

farm succession and household size. The collected farm-holding information includes

the farm size (agricultural area), the proportion of rented land, and size of livestock.

Risk management strategies

Finally, the survey included a list of 16 risk management tools (see Appendix 3.A

Table 3.A.2). The list was developed following earlier research focusing on similar

farm types and/or similar production regions (e.g. Meuwissen, Huirne, and Hardaker,

25

However, the scale does not allow for a quantitative interpretation in terms of a risk aversion

coefficient.


93

2001, Schaper, Bronsema, and Theuvsen, 2012, Musser and Patrick, 2002) as well as

in-depth expert interviews with extension service consultants and farmers. These

interviews also clearly revealed that it is not a single risk management strategy which

is relevant for our case study. In contrast, a combination of strategies is usually

applied. Consequentially, we included a list of 16 different risk management strategies

in the survey, asking farmers to choose those strategies applied on their farm. For the

subsequent analysis, we classify the 16 risk management strategies according to three

different areas where the farmer may shift her/his resources to (following van der

Ploeg and Roep, 2003). First, resources are kept on the farm holding, focusing on

agricultural production, second, the resources are shifted towards risk management

efforts on the farm not primarily related to agricultural production. Third, resources are

shifted away from the farm holding to a third party. In the remainder of this paper, the

first category will be referred to as “on-farm agricultural” and includes the following

risk management strategies: investment in new technologies (e.g. irrigation),

agricultural diversification, risk adapted production and use of robust or resistant

varieties and breeds. The second category named “on-farm non-agricultural” includes:

non-agricultural diversification, work harder or cut private expenses, cooperation with

other farmers and building reserves. Finally, the third category labeled “off-farm” risk

management strategies includes: working off the farm, yield insurance, hail insurance,

multiple risk insurance, business liability insurance + floor-coverage + environmental

liability, legal protection insurance, trading on the commodity futures exchange, and

off-farm investment. For the further analysis, we categorize the farmers’ choices of

3.4 Methodology

94

risk management. As all farmers use a portfolio of different risk management tools, we

categorize them by maximizing the mean over all choices in each risk management

category (i.e. they are categorized depending on the largest share of risk management

tools applied).

3.4 Methodology

To identify the determinants of risk management behavior we estimate a multinomial

probit model (Verbeek, 2008). This choice (e.g. in favor of a multinomial logit model)

is motivated by the fact the unobserved error terms of the separate probit models are

very likely not independent26

. Ignoring this correlation in analyzing the simultaneous

adoption of risk management tools may lead to biased estimates of the choice

probabilities and incorrect estimates of the standard errors of the parameters (Verbeek,

2008). We estimate the observed choices as a function of risk preferences, perceptions

and socio- economic farm, farmer and household characteristics (a detailed description

of all variables is found in Table 3.2). We simultaneously estimate the farmers’

preferred risk management category: on-farm agriculture (j=1), on-farm non-

agriculture (j=2) and off-farm and mixed strategies27

(j=0). In other words, we assume

the farmer to prefer one of the three categories depending on the averagely highest

amount of risk management tools chosen in each category (j=1 if the farmer choses on

26

When testing the multinomial logit model for our data, the seemingly unrelated estimation

test indicates a violation of the assumption of independent alternatives (Weesie, 1999). The

results of the estimation using the multinomial logit models can be found in Appendix 3.A

Table 3.A.5. 27

Seventeen farmers use a mix of strategies with no clear favored category; they are grouped in

the off-farm category as the preferred mix always includes off-farm strategies.

3.4 Methodology

95

average mainly on-farm agriculture related strategies; j=2 if the farmer choses on

average mainly on-farm non- agriculture related strategies and j=0 if the farmer choses

on average mainly off-farm strategies to manage her/his risks). Specifically, we

estimate the probability that one of the risk management classes is preferred (i.e. is

chosen more frequently) over the other classes:

𝑦𝑖𝑗∗ = 𝛽𝑖𝑗𝑥𝑗 + 𝜀𝑖𝑗 𝜀𝑖𝑗 ~ 𝑁 (0, 𝛴) and 𝑗 = (0,1,2)

with 𝑦𝑖 = {

1 if 𝑦𝑖 𝑜𝑛 𝑓𝑎𝑟𝑚 𝑎𝑔𝑟𝑖𝑐𝑢𝑙𝑡𝑢𝑟𝑒∗ > 0

2 if 𝑦𝑖 𝑜𝑛 𝑓𝑎𝑟𝑚 𝑛𝑜𝑛−𝑎𝑔𝑟𝑖𝑐𝑢𝑙𝑡𝑢𝑟𝑒∗ > 0

0 otherwise.

𝑥𝑗 = vector of observable farm, farmer and household characteristics

Where 𝛽𝑖𝑗 is a vector of parameters specific to the j-th alternative associated with the

vector 𝑥𝑗, which contains the observable farm, farmer and household characteristics.

The error terms 𝜀𝑖𝑗 are assumed to be multivariate normally distributed with mean

zero. The category off-farm risk management is chosen as base category.

The estimation is repeated for each of the six risk preference elicitation

methods included plus the average over four business statements (MPL, SA, BS

production, BS market and prices, BS financial, BS agriculture generally, BS average)

in order to test which risk preference elicitation method relates best to the farmers’ risk

management choice using the statistical software STATA13.1.

In addition, we assess the cross-method consistency of risk preference

elicitation methods in our analysis, by testing whether risk preferences elicited with

individual approaches point in the same direction. To account for the ordinal nature of

risk preferences, tests on rank correlations are used.

3.5 Data

96

3.5 Data

In Table 3.2 we present a description of all variables included in the further analysis.

Table 3.3 provides summary statistics for all variables used. We include only

observations with no missing values in order to have a consistent data set throughout

the different estimations28

. Consequently, the data set reduces to 56 farmers. We find

only two missing values for the MPL and one for the SA task, thus there is great

evidence that the instructions of the MPL, combined with the visual aid decreased

complexity. Furthermore, the inconsistency rate in the MPL is low (10% of farmers

switch multiple times in the MPL). Following Holt and Laury (2002), and Abdellaoui,

Driouchi, and L’Haridon (2011), we argue that the bias regarding the average number

of safe choices is negligible, calculating the corresponding CRRA interval mid-point

based on the individual’s first transition to the riskier choice B. On average across all

surveyed farms, the main farm operator are 45 years old, cultivate about 107 ha of land

of which 50% are rented. Typically for the area most farms are livestock oriented (pig

or cattle), with an average of 951 pigs or piglets and 73 cattle29

. The regional average

for agricultural area and livestock is around 20% lower; this is very likely because we

have an overrepresentation of full time farmers in our survey (none of the participants

earning more than 50% of their income outside agriculture). We find that the sample

28

Note that this listwise deletion of missing values ensures comparability of the models.

Furthermore, we do not find estimation biases, i.e. can assume that missing values are

random. 29

The full data set and estimation code are available in the supplemental file attached.

3.5 Data

97

contains highly educated30

and slightly optimistic farmers, who self-assess their

numeracy to be good. They attend on average two agricultural training workshops per

year. Farmers in our sample perceive market and price risks most severely, followed by

political and structural risks. This is in line with findings amongst Dutch livestock

farmers by Meuwissen, Huirne, and Hardaker (2001) as well as results gained in a

study by Flaten et al. (2005) for Norwegian dairy farmers. Most of the farmers (80%)

claim to have experienced major losses over the past five years. Furthermore, most

farm businesses are operated with 2.5 labor units and on average five people live in the

farm household. The farm succession is for most farmers planned and quite certain.

30

The mode degree of education in the sample is at the level of state certified agriculturalist

(implying a minimum of 13 years of schooling).

3.5 Data

98

Table 3.2: Variable description

Variable Variable definition

Farmer

characteristics

Risk preferences

MPL CRRA interval mid-point

SA General risk preferences on inverse scale from 0 (= very unwilling to take

risks) to 10 (= very willing to take risks)

Willingness to take more risks than my colleagues with respect to…

BSProd production, on scale from 1 = agree, 5 = don't agree

BSMark marketing and prices production, on scale from 1 = agree, 5 = don't

agree

BSFin financial issues, on scale from 1 = agree, 5 = don't agree

BSAg farming in general, on scale from 1 = agree, 5 = don't agree

Ø BS Average of four BS

Age Years

Educ Highest degree of education according to the German schooling system (0-

10)

Optim Life satisfaction in one year (on a scale from 1 = “not satisfied” to 10 =

“very satisfied”) - life satisfaction now (on a scale from 1 = “not satisfied”

to 10 = “very satisfied”)

SN Mean subjective numeracy score (1 = very good numeracy; 6 = very bad

numeracy)

AgricTrain Attendance of agricultural training workshops per year (0 = none; 1 = one;

2 = two to three; 4 = four to five; 4 = more than five)

Risk perception Perceived probability of occurrence (1 = “not likely”, 5 = “very likely”)

multiplied by perceived impact on 1 to 5 scale (1 = “no impact”, 5 = “very

strong impact”) in four domains

PerMarkRisk Perceived market risks

PerPolRisks Perceived political risks

PerProdRisk Perceived production risks

PerFinRisk Perceived financial risks

PerOtherRisk Perceived other risks

ExpLosses 1 if experienced severe losses over the past five years

Household

characteristics

WF Number of full time workforce available

FarmSuc 1 if succession is planned and sure, 0.5 if succession is planned and quite

sure, 0 if succession is not planned in the next 15 years, -0.5 if succession

is quite unsure, -1 if succession is unsure

HHS Household members living in the farm household

Farm

characteristics

AgricArea Agricultural area in ha

PropRentLand % of rented land that is cultivated

LivePig Number of pigs or piglets

LiveCattle Number of cattle

3.5 Data

99

Table 3.3: Summary statistics of complete surveys N = 56

All farmers N = 56 Off-farm N = 21 On-farm agric N = 16 On-farm non-agric N = 19

Variable Mean SD Mode Mean SD Mode Mean SD Mode Mean SD Mode

Farmer

characteristics

MPL 0.30 0.66 0.54 0.28 0.68 0.54 0.44 0.50 0.54 0.20 0.78 0.28

SA 5.45 2.19 6.00 5.38 2.25 6.00 5.44 2.45 6.50 5.53 2.01 6.00

BSProd 3.02 1.10 3.00 3.05 1.16 3.00 2.81 1.11 2.50 3.16 1.07 3.00

BSMark 3.125 1.03 3.00 3.10 1.18 3.00 2.94 1.00 3.00 3.32 0.89 3.00

BSFin 3.18 1.21 3.00 2.95 1.28 3.00 3.56 1.12 4.00 3.11 1.15 3.00

BSAg 3.02 1.07 3.00 3.05 1.20 3.00 2.75 1.07 2.50 3.21 0.92 3.00

BSProd 3.09 0.94 3.00 3.04 1.09 3.00 3.02 0.88 2.88 3.20 0.85 3.00

Ø BS 3.02 1.10 3.00 3.05 1.16 3.00 2.81 1.11 2.50 3.16 1.07 3.00

Age 45.29 11.16 47.00 44.67 11.49 47.00 46.38 12.71 47.50 45.05 9.88 47.00

Educ 8.32 1.39 8.00 8.48 0.87 8.00 8.38 1.63 8.00 8.11 1.66 8.00

Optim 0.18 0.74 0.00 0.24 0.70 0.00 0.13 0.50 0.00 0.16 0.96 0.00

SN 2.06 0.60 2.07 1.87 0.74 1.71 2.32 0.43 2.29 2.05 0.50 2.14

AgricTrain 2.79 1.09 3.00 2.95 0.97 3.00 2.69 0.95 2.50 2.68 1.34 3.00

PerMarkRisk 13.53 3.85 13.80 14.23 3.11 13.80 12.47 4.58 12.95 13.66 3.93 14.00

PerPolRisks 12.42 3.49 12.69 12.44 3.58 11.50 12.25 3.24 13.13 12.53 3.78 12.63

PerProdRisk 10.51 3.04 10.20 10.53 9.82 10.20 10.32 2.92 10.30 10.65 3.19 10.20

PerFinRisk 9.62 4.18 9.50 9.74 4.36 9.00 8.69 4.71 7.00 10.26 3.54 10.50

PerOtherRisk 9.86 2.74 9.60 9.55 2.44 9.60 10.38 3.09 10.30 9.77 2.84 9.40

ExpLosses 0.82 0.39 1.00 0.71 0.46 1.00 0.94 0.25 1.00 0.84 0.37 1.00

Household

characteristics

WF 2.52 1.23 2.25 2.58 1.30 3.00 2.39 1.01 2.25 2.56 1.37 2.00

FarmSuc 0.55 0.53 0.50 0.50 0.55 0.50 0.47 0.59 0.50 0.66 0.44 1.00

HHS 4.98 1.69 5.00 4.95 1.75 5.00 4.94 2.02 6 5.05 1.39 5.00

Farm

characteristics

AgricArea 106.97 91.79 89.00 101.79 54.58 90.00 90.84 47.96 83.50 126.28 140.92 87.00

PropRentLand 0.29 0.50 0.52 0.59 0.30 0.67 0.46 0.26 0.50 0.41 0.27 0.46

LivePig 950.64 1159.68 378.50 691.38 1024.57 0.00 702.31 1154.94 0.00 1446.3

2 1199.48 1700.00

LiveCattle 72.73 117.34 0.00 93.51 101.48 100.00 61.56 88.16 0.00 59.11 152.77 0.00

3.6 Results

100

3.6 Results

All risk aversion coefficients elicited using different risk preference elicitation methods

show that farmers in our sample are – on average – risk averse. However, there is

heterogeneity with respect to risk preferences within the overall sample. The average

CRRA interval mid-point is 0.30 (Table 3.2), which corresponds to values elicited by

Maart-Noelck and Musshoff (2013) for German farmers. However, in the self-

assessment of risk preferences are farmers in our sample on average more risk averse

(5.45) than German farmer in the sample of Maart-Noelck and Musshoff (2013) (4.9).

Based on the four business statements, we find most farmers to identify a relative risk

neutral position in all four relevant domains (mode = 3.00), which is in line with

results of Meuwissen, Huirne, and Hardaker (2001) in a sample of Dutch livestock

farms. The illustrated differences in means is additionally tested significantly using

Hottelings T-squared test. Additionally, Appendix 3.A Figure 3.A.2 shows the Kernel

density plots of CRRA estimates, self-assessment and average business statements in

the sample, as well as risk neutrality (dotted line).

To analyze how risk attitude parameters elicited using different risk preference

elicitation methods pertain farmers’ risk management decisions, we first examine

farmers’ risk preference consistency across the three methods. Table 3.4 shows the

Spearman correlation coefficient of risk preferences elicited through the three different

methods31

(generating seven parameters of risk aversion). We find a significantly

31

Additionally Pearson correlation coefficients are found in the Appendix 3.A Table 3.A.3.

3.6 Results

101

positive correlation of all preference estimations. Thus, all risk preference elicitation

methods reveal a consistent representation of the farmers’ risk preferences (i.e. if risk

averse in one task most farmers also behave risk averse in the other tasks).

Table 3.4: Spearman correlation coefficients of risk preferences elicited with different methods

N=6132

MPL SA Ø BS BSProd BSMark BSFin BSAg

MPL 1.000

SA 0.714*** 1.000

Ø BS 0.417*** 0.661*** 1.000

BSProd 0.492*** 0.675*** 0.844*** 1.000

BSMark 0.357*** 0.491*** 0.804*** 0.677*** 1.000

BSFin 0.264** 0.521*** 0.753*** 0.492*** 0.428*** 1.000

BSAg 0.349*** 0.586*** 0.905*** 0.740*** 0.701*** 0.602*** 1.000

** Significant at the 5% level, *** significant at the 1% level

Additionally, we checked for multicollinearity for the variables used to test for

risk attitude and risk perception. Contrarily to suggestions by some researchers

(Menapace, Colson, and Raffaelli, 2013, van Winsen et al., 2014), we do not find a

significant correlation of risk attitude and risk perception in our sample.

Table 3.5 and Table 3.6 summarize the estimated coefficients and standard

errors for seven multinomial probit models estimated. The independent variables,

which are described in Table 3.2, are equivalent across the seven models except for the

specification of the risk aversion variable, which varies in each model. More

specifically we estimate separate models for each risk preference elicitation method:

32

Note that we excluded three non-responding farmers for this analysis.

3.6 Results

102

MPL, SA, BS production, BS marketing and prices, BS finances, BS agriculture

generally and the average over all business statements. We find 21 of the participants

have on average mostly off-farm risk management strategies, 16 choose on average

mainly on-farm agriculture related risk management tools and 19 engage mainly in on-

farm non-agriculture related risk management strategies. For all multinomial probit

models the hypothesis of identical probabilities for each category can be rejected

(Table 3.5 and Table 3.6).

We find that greater risk aversion increases the probability that farmer’s focus

on on-farm strategies (compared to choosing off-farm strategies). More specifically,

when risk preferences are measured via the self-assessment (SA), business statement

(BS) on production risks, marketing and price risks, and agricultural risks in general, a

positive impact of risk aversion on on-farm non-agriculture strategies is found.

However, for risk preferences elicited using the contextualized MPL and in the

business statement on financial risks the probability to mainly rely on on-farm

agriculture related risk management tools increases.

Furthermore, our results show that with increasing age farmers are more likely

to engage in on-farm non-agriculture related risk management tools compared to off-

farm strategies. In contrast, older farmers are less likely choose on-farm agriculture

related risk management strategies in two out of our seven models. Education has a

positive effect on the probability of a larger share of on-farm agriculture related risk

management tools compared to off-farm strategies. Risk literacy has a significant

effect on the probability of farmers to focus on on-farm risk management. More

3.6 Results

103

specifically, with increasing risk literacy the probability of farmers to engage in more

on-farm risk management strategies decreases. In contrast, more risk literate farmers

are more likely to use off-farm measures. A high perception of market risks decreases

the probability of farmers to focus on on-farm agricultural risk management strategies

compared to off-farm strategies. A higher perception of other risks (i.e. workforce,

societal acceptance) increases the probability of farmers focusing on on-farm

agriculture related risk management strategies in five out of seven of the estimated

models. Furthermore, larger farms are more likely to engage in off-farm risk

management strategies. We find an increasing share of rented land is associated with a

higher probability of farmers focusing on off-farm risk management tools than on-farm

non-agriculture related tools. Farmers that experienced losses in any of the five

domains show an increased probability to focus on on-farm risk management tools in

their risk management portfolio. If the farm succession is sure the probability to

engage in on-farm not agriculture related risk management strategies increases.

3.6 Results

104

Table 3.5: Multinomial probit estimates for MPL, SA and average BS

MPL SA Ø BS

N = 56 On-farm

agriculture

On-farm

non-

agriculture

On-farm

agriculture

On-farm

non-

agriculture

On-farm

agriculture

On-farm

non-

agriculture

Farmer

characteristics

Risk aversion 1.870**

(0.811)

0.229

(0.542)

-0.047

(0.227)

0.328*

(0.201)

0.916

(0.630)

1.274**

(0.570)

Age -1.124**

(0.552)

0.602*

(0.330)

-1.020**

(0.468)

0.661*

(0.367)

-1.163**

(0.511)

0.241

(0.358)

Age^2 0.014**

(0.006)

-0.006*

(0.004)

0.012**

(0.005)

-0.007*

(0.004)

0.014**

(0.006)

-0.003

(0.004)

Educ 0.531

(0.338)

-0.300

(0.287)

0.982**

(0.457)

-0.337

(0.340)

0.856**

(0.393)

-0.544

(0.379)

Optim -0.268

(0.499)

-0.394

(0.444)

-0.099

(0.511)

-0.260

(0.450)

0.342

(0.486)

0.067

(0.411)

SN 3.578***

(1.158)

1.424*

(0.645)

3.409***

(1.023)

1.264**

(0.649)

3.447***

(0.988)

1.274**

(0.551)

AgricTrain -0.125

(0.365)

0.323

(0.326)

-0.243

(0.427)

0.002

(0.361)

-0.432

(0.407)

0.143

(0.311)

PerMarkRisk -1.067***

(0.326)

-0.211

(0.139)

-1.049***

(0.310)

-0.306*

(0.157)

-1.148***

(0.337)

-0.339**

(0.158)

PerPolRisks 0.131

(0.186)

0.119

(0.136)

0.243

(0.183)

0.100

(0.147)

0.194

(0.179)

0.109

(0.142)

PerProdRisk -0.126

(0.186)

-0.068

(0.134)

-0.089

(0.147)

-0.092

(0.134)

-0.117

(0.147)

-0.067

(0.136)

PerFinRisk -0.151

(0.115)

0.041

(0.093)

-0.261*

(0.137)

0.106

(0.100)

-0.169

(0.141)

0.126

(0.116)

PerOtherRisk 0.400**

(0.220)

0.030

(0.171)

0.363*

(0.220)

0.035

(0.165)

0.426*

(0.221)

0.035

(0.165)

ExpLosses 10.537***

(2.683)

2.524**

(0.957)

9.968***

(2.294)

2.391***

(0.896)

9.877***

(2.195)

2.577***

(0.943)

Household

characteristics

WF -0.020

(0.435)

-0.015

(0.463)

-0.015

(0.463)

0.042

(0.366)

0.176

(0.440)

0.375

(0.341)

FarmSuc -0.750

(1.260)

2.172**

(0.962)

-1.311

(1.396)

2.363**

(0.977)

-0.945

(1.117)

2.159***

(0.813)

HHS -0.485

(0.340)

0.136

(0.232)

-0.357

(0.287)

0.089

(0.244)

-0.354

(0.316)

0.123

(0.247)

Farm

characteristics

AgricArea -0.022*

(0.012)

0.003

(0.004)

-0.020

(0.013)

0.003

(0.004)

-0.024*

(0.012)

0.001

(0.004)

PropRentLand 1.530

(2.339)

-5.549***

(1.664)

1.471

(2.386)

-4.955***

(1.500)

1.107

(2.505)

-5.915***

(1.746)

LivePig 0.001

(0.001)

0.000

(0.000)

0.000

(0.001)

0.001**

(0.000)

0.001

(0.001)

0.001***

(0.000)

LiveCattle 0.011*

(0.005)

0.001

(0.004)

0.008**

(0.004)

0.003

(0.003)

0.010**

(0.005)

0.005

(0.003)

Constant 13.758

(11.152)

-15.693*

(8.232)

8.127

(9.282)

-12.190

(8.668)

10.227

(9.535)

-8.804

(8.519)

Wald Chi2(40) 95.40 90.08 97.48

Log

pseudolikelihood -33.728 -34.160 -33.963

Prob > chi2 = 0.000 0.000 0.000

* Significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level

3.6 Results

105

Table 3.6: Multinomial probit estimates for BS production, BS marketing and prices, BS

finances and BS agriculture generally

BSProd BSMark BSFin BSAg

N = 56 On-farm

agriculture

On-farm

non-

agriculture

On-farm

agriculture

On-farm

non-

agriculture

On-farm

agriculture

On-farm

non-

agriculture

On-farm

agriculture

On-farm

non-

agriculture

Farmer

characteristics

Risk aversion 0.519

(0.525)

0.546**

(0.370)

0.449

(0.502)

1.018**

(0.440)

1.118**

(0.445)

0.556

(0.408)

-0.489

(0.535)

0.718*

(0.410)

Age -1.067**

(0.518)

0.430

(0.308)

-1.026**

(0.460)

0.282

(0.314)

-1.331***

(0.466)

0.375

(0.388)

-1.200**

(0.533)

0.259

(0.379)

Age^2 0.013**

(0.006)

-0.005

(0.003)

0.013**

(0.005)

-0.003

(0.003)

0.016***

(0.005)

-0.004

(0.004)

0.015**

(0.006)

-0.003

(0.004)

Educ 0.991**

(0.414)

-0.279

(0.282)

0.852**

(0.390)

-0.511

(0.342)

0.802*

(0.434)

-0.379

(0.368)

1.156***

(0.407)

-0.342

(0.297)

Optim 0.169

(0.447)

-0.233

(0.413)

0.166

(0.497)

-0.081

(0.408)

0.891*

(0.474)

0.037

(0.520)

-0.124

(0.703)

-0.174

(0.415)

SN 3.143***

(1.019)

1.211**

(0.559)

3.411***

(0.955)

1.510***

(0.543)

3.230***

(0.997)

1.365**

(0.607)

3.923***

(1.030)

1.115**

(0.502)

AgricTrain -0.403

(0.407)

0.187

(0.309)

-0.446

(0.407)

-0.026

(0.313)

-0.677*

(0.389)

0.152

(0.319)

-0.052

(0.423)

0.221

(0.285)

PerMarkRisk -1.017***

(0.302)

-0.234

(0.131)

-1.085***

(0.336)

-0.299*

(0.159)

-1.192***

(0.276)

-0.344**

(0.160)

-1.104***

(0.310)

-0.181

(0.136)

PerPolRisks 0.153

(0.171)

0.083

(0.142)

0.204

(0.183)

0.099

(0.143)

0.274

(0.168)

0.204

(0.139)

0.286

(0.185)

0.078

(0.136)

PerProdRisk -0.102

(0.143)

-0.050

(0.128)

-0.055

(0.147)

-0.006

(0.142)

-0.162

(0.143)

-0.092

(0.139)

-0.076

(0.149)

-0.035

(0.138)

PerFinRisk -0.216

(0.136)

0.057

(0.094)

-0.222

(0.139)

0.058

(0.095)

-0.100

(0.145)

0.114

(0.122)

-0.275*

(0.144)

0.054

(0.085)

PerOtherRisk 0.444**

(0.224)

0.060

(0.154)

0.391*

(0.220)

0.026

(0.166)

0.333

(0.208)

-0.062

(0.176)

0.291

(0.244)

-0.044

(0.162)

ExpLosses 9.451***

(2.136)

2.257***

(0.869)

9.837***

(2.169)

2.454***

(0.927)

9.472***

(2.100)

2.344**

(0.904)

11.146***

(2.356)

2.434***

(0.911)

Household

characteristics

WF -0.014

(0.429)

0.236

(0.360)

0.119

(0.490)

0.360

(0.347)

-0.168

(0.422)

0.272

(0.367)

-0.123

(0.542)

0.211

(0.350)

FarmSuc -0.895

(1.169)

2.183**

(0.900)

-1.013

(1.204)

1.913**

(0.830)

-0.625

(1.069)

2.055**

(0.929)

-1.835

(1.203)

1.780**

(0.817)

HHS -0.317

(0.311)

0.120

(0.238)

-0.403

(0.309)

0.059

(0.229)

-0.387

(0.312)

0.122

(0.266)

-0.333

(0.305)

0.106

(0.258)

Farm

characteristics

AgricArea -0.021

(0.014)

0.001

(0.004)

-0.021*

(0.012)

0.001

(0.004)

-0.019

(0.012)

0.003

(0.004)

-0.020

(0.012)

0.002

(0.004)

PropRentLand 1.934

(2.620)

-4.709***

(1.449)

1.163

(2.490)

-5.603***

(1.628)

1.368

(2.413)

-5.559***

(1.654)

1.155

(2.449)

-5.273***

(1.566)

LivePig 0.001

(0.001)

0.001**

(0.000)

0.001

(0.001)

0.001***

(0.000)

0.001

(0.001)

0.001**

(0.000)

0.000

(0.001)

0.001**

(0.000)

LiveCattle 0.010*

(0.005)

0.003

(0.003)

0.009**

(0.004)

0.005

(0.003)

0.013***

(0.005)

0.003

(0.003)

0.008

(0.005)

0.004

(0.003)

Constant 7.924

(8.896)

-13.183*

(7.630)

8.573

(9.335)

-9.132

(7.984)

15.727*

(9.123)

-9.788

(9.578)

10.602

(10.712)

-8.911

(8.247)

Wald Chi2(40) 128.59 108.23 124.31 110.47

Log

pseudolikelihood -35.046 -33.783 -33.557 -33.441

Prob > chi2 = 0.000 0.000 0.000 0.000


3.7 Discussion

106

3.7 Discussion

Risk preferences derived from three different elicitation methods used in our analysis

are all highly correlated. Thus, our analysis shows that high between method

consistency can be obtained by using a contextualization of the MPL combined with a

visual aid. Consequentially, our findings point towards possible improvements to

earlier studies on farmers’ risk preferences that reveal high between method

inconsistencies for non-contextualized MPL settings (e.g. Reynaud and Couture, 2012,

Anderson and Mellor, 2009, Menapace, Colson, and Raffaelli, 2015).

In addition, our results show that risk averse farmers are more likely to prefer a

larger share of on-farm risk management tools (compared to off-farm strategies). This

confirms Flaten et al. (2005), Hellerstein, Higgins, and Horowitz (2013), and

Menapace, Colson, and Raffaelli (2015) who find that risk averse farmers are less

likely to apply market based risk management strategies such as crop insurance.

Hellerstein, Higgins, and Horowitz (2013) argue similar to Herberich and List (2012)

that this counter intuitive result stems from the background risk influencing the

farmers’ decision in an experimental setting, and call for a richer structural model of

farming practices and experimental choices. By contextualizing the MPL and including

a large portfolio of risk management strategies, we attempt to eliminate some of the

background risk. Our results show that for some off-farm strategies included in our

studies (e.g. trading on the commodity futures exchange) the argument made by

Holden and Quiggin (2016) that greater risk aversion is associated with lower adoption

rates of new technologies due to the uncertainty associated with the new risk

3.7 Discussion

107

management strategy. Moreover, farming is usually associated with a high degree of

identification of farmers with their profession and farm holding. Thus, many farmer

may evaluate their yield and farming business as irreplaceable in the sense that there

are no equivalent commodities available on the market (Cook and Graham, 1977). For

a risk averse farmer this may imply risk-taking behavior that is reduced by risk

aversion (resulting in on-farm risk management strategies) and a reduced demand of

insurance.

Furthermore, our results show domain dependence of risk preferences

(Dohmen et al., 2011). Farmers showing risk averse preference in the financial domain

(MPL and financial business statement) are more likely to focus on on-farm

agricultural risk management strategies, we can hypothesis that they have less trust in

financial markets and thus focus on on-farm solutions. Farmers who are less willing to

take risks with respect to agricultural production, marketing and pricing and agriculture

in general than their colleagues are more likely to focus on on-farm non-agricultural

risk management strategies, i.e. keeping their resources on the farm but away from

agricultural production.

Furthermore, our findings show that off-farm risk management is preferred by

middle aged farmers (between 40 and 50), whereas younger farmers prefer on-farm

agriculture related risk management tools and older farmers prefer on-farm non-

agriculture related risk management tools. This result confirms findings by van Winsen

et al. (2014) and Flaten et al. (2005), who find that older farmers are less likely to

apply off-farm risk management strategies. Consistent with Potter and Lobley (1996)

3.7 Discussion

108

who describe the farm family development cycle as periods of excess and undersupply

of resources our results show that younger farmers can be assumed to have excess

labor capacities and lack of financial resources, thus agricultural production related on-

farm risk management tools are dominant in their risk management portfolio. With

increasing age, farmers’ focus of risk management strategies shifts towards off-farm

tools. This effect however saturates and in the final stage of the professional career, on-

farm non-agriculture related risk management tools are preferred. Older farmers are

concerned with building up a future for the next generation on the farm and thus shift

the excess resources towards new on-farm ventures or building reserves. More general,

our results show that building up long-term ties for the future generation plays a

significant role in the choice of risk management tools. Our findings show that farm

households with anticipation of succession are more likely to focus on on-farm non-

agriculture related risk management tools. Potter and Lobley (1996) explain that farm

households with successors focus on setting up capital to provide a living for the

successor. Thus, depending on the stage of the succession process these farms are more

likely to focus their risk management strategies on building reserves, working harder,

cooperating with other farmers and investing in on-farm non-agricultural

diversification ventures.

Farmers that have a higher subjective numeracy score are more likely to use

more off-farm risk management tools. As subjective numeracy has been shown to

correlate positively with risk comprehension (Zikmund-Fisher et al., 2007), this results

shows that farmers with a better understanding of probabilities and preferences for

3.7 Discussion

109

numbers are more likely to take on more insurances, trade on the commodity futures

exchange, or engage in off-farm work and investments i.e. activities that require high

numeracy skills. Moreover, we find that risk perception is directly influencing the

decision of risk management strategies applied. Thus, we can confirm findings by

Meuwissen, Huirne, and Hardaker (2001) and Flaten et al. (2005) who find a

significant influence of risk perception on farmers risk behavior. More specifically, a

high perception of market risks as well as other risks increase the probability of

farmers to focus on off-farm risk management strategies. Our results show that with

increasing farm size, the likelihood to apply on-farm risk management strategies

related to agricultural production decreases. Farmers that experienced major losses in

the past 5 years are more likely to engage in on-farm (both agricultural and non-

agricultural) risk management strategies, compared to off-farm risk management tools.

This result is counter intuitive as experienced past losses are assumed to increase risk

perception and in turn the probability to shift capital and labor outside the farming

business (Menapace, Colson, and Raffaelli, 2015). Our finding indicates that better

instruments or better communication for off-farm risk management tools might be

needed to allow farmers to diversify also outside of risky on-farm activities.

An increasing acreage of the farm is associated with greater spatial dispersion,

likely indicating larger diffusion of the location of farmland, implying that farmers

have already reduced some production risk. Our results show that these farmers are

less likely to engage in a greater share of on-farm risk management strategies related to

agricultural production compared to off-farm strategies. Furthermore, we find that an

3.8 Conclusion

110

increasing proportion of rented land increases the probability to apply more off-farm

risk management strategies compared to on-farm non-agriculture related tools. Thus,

we can confirm findings by Velandia et al. (2009) and Mishra and El-Osta (2002) who

claim that a higher proportion of rented land is associated with higher risk exposure,

lower wealth, means, and incentives to build-up long-term capacities for risk bearing

on the farm, resulting in a greater need to spread the risk to a third party.

3.8 Conclusion

In this article, we have contrasted three alternative hypothetical methods for assessing

risk preferences that vary in terms of their simplicity and contextual framing and

payoff scale. We find that risk preferences are context depending and by framing the

widely used MPL in an agricultural context we found significant evidence for it to

pertain well to real decision making. Farmers have a number of options in managing

agricultural risks and many of them utilize multiple risk management tools

simultaneously. However, most literature on factors affecting adoption of multiple risk

management tools has not addressed this aspect. It is often implicitly assumed that the

decision to adopt one risk management tool is independent of the decision to adopt

other risk management tools. In contrast, we show that focusing on risk management

portfolios, going beyond single risk management tools, is required. We find that risk

averse farmers are more likely to prioritize on-farm risk management strategies over

off-farm strategies. Our analysis shows that counter intuitively risk averse farmers are

less likely to choose off-farm risk management strategies. Explanations can be found in

3.8 Conclusion

111

the novel and innovative nature of some off-farm risk management strategies included,

as well as in the fact that for most farmers the farm business (or yield) is considered as

in irreplaceable commodity that cannot be valued in marketable terms. In addition, our

findings contribute to solve this puzzle by showing that risk averse farmers do not

choose no risk management as an alternative but focus on on-farm measures. Further

research should abstain from focusing on the analysis of single risk management tools

but further develop a holistic approach including the whole portfolio of risk

management tools applied by farmers. As we collected the data analyzed in this article

via a self-deducted paper pencil survey, the information collected is limited to the main

farm operator. Nevertheless, there is evidence that the farming couple is deciding on

the household risk management strategy jointly (see e.g. Benjamin and Kimhi, 2006).

Consequently, there might be additional factors influencing the farmer’s risk

management choice not considered in this analysis (e.g. demographic characteristics of

the farmer’s spouse including risk preferences, occupation, age etc.). Furthermore,

there are additional factors influencing the farmer’s risk management choice we were

not able to collect in the survey due to the high sensitivity of information in the

specific context of the case study area (e.g. exact location or income). Moreover, we

find that age, risk perception, subjective numeracy, farm succession, farm size and the

proportion of rented land play a role when explaining farmers’ risk behavior. Extension

educators and other risk management information providers in the survey area may be

able to tailor their programs better, based on the results gained from this analysis.

3.9 References

112

3.9 References

Abdellaoui, Mohammed, Ahmed Driouchi, and Olivier L’Haridon. 2011. "Risk

aversion elicitation: reconciling tractability and bias minimization." Theory and

Decision 71 (1):63-80.




Benjamin, Catherine, and Ayal Kimhi. 2006. "Farm work, off-farm work, and hired

farm labour: estimating a discrete-choice model of French farm couples' labour

decisions." European Review of Agricultural Economics 33 (2):149-171.

Bishu, Kinfe G., Seamus O’Reilly, Edward Lahiff, and Bodo Steiner. 2016. "Cattle

farmers’ perceptions of risk and risk management strategies: evidence from

Northern Ethiopia." Journal of Risk Research:1-20.

Bommier, Antoine, and François Le Grand. 2014. "Too Risk Averse to Purchase

Insurance? A theoretical glance at the annuity puzzle." Journal of Risk and


Bougherara, Douadia, Xavier Gassmann, and Laurent Piet. 2011. "A structural

estimation of French farmers’ risk preferences: an artefactual field experiment."

Working Paper SMART – LERECO 11 (06).



Cook, Philip J., and Daniel A. Graham. 1977. "The Demand for Insurance and

Protection: The Case of Irreplaceable Commodities." The Quarterly Journal of

Economics 91 (1):143-156.

Crosetto, Paolo, and Antonio Filippin. 2013. "The “bomb” risk elicitation task."


de Mey, Yann, Erwin Wauters, Dierk Schmid, Markus Lips, Mark Vancauteren, and

Steven van Passel. 2016. "Farm household risk balancing: empirical evidence from

Switzerland." European Review of Agricultural Economics 43 (4).




(3):522–550.

3.9 References

113

Eckel, Catherine C., and Philip J. Grossman. 2008. "Forecasting risk attitudes: An

experimental study using actual and forecast gamble choices." Journal of

Economic Behavior & Organization 68 (1):1-17.

Fagerlin, Angela, Brian J. Zikmund-Fisher, Peter A. Ubel, Aleksandra Jankovic, Holly

A. Derry, and Dylan M. Smith. 2007. "Measuring numeracy without a math test:

development of the Subjective Numeracy Scale." Medical Decision Making 27

(5):672–680.

Finger, Robert, and Niklaus Lehmann. 2012. "The influence of direct payments on

farmers’ hail insurance decisions." Agricultural Economics 43 (3):343-354.

Flaten, Ola, Gudbrand Lien, Matthias Koesling, Paul Steinar Valle, and Martha

Ebbesvik. 2005. "Comparing risk perceptions and risk management in organic and

conventional dairy farming: empirical results from Norway." Livestock Production

Science 95 (1):11–25.

Gneezy, Uri, and Jan Potters. 1997. "An Experiment on Risk Taking and Evaluation

Periods." The Quarterly Journal of Economics 112 (2):631-645.

Hansson, Helena, and Carl Johan Lagerkvist. 2012. "Measuring farmers’ preferences

for risk: a domain-specific risk preference scale." Journal of Risk Research 15

(7):737-753.

Harrison, Glenn W., Morten I. Lau, and Elisabet E. Rutström. 2007. "Estimating risk

attitudes in denmark: A field experiment." The Scandinavian Journal of Economics

109 (2):341-368.

Harrison, Glenn W., John A. List, and Charles Towe. 2007. "Naturally occurring

preferences and exogenous laboratory experiments: A case study of risk aversion."

Econometrica: Journal of the Econometric Society 75 (2):433-458.

Heinemann, Frank. 2008. "Measuring risk aversion and the wealth effect." In Risk

aversion in experiments, edited by James C. Cox and Glenn W. Harrison, 293-313.

online: 08 Mar 2015: Emerald Group Publishing Limited.




Herberich, David H., and John A. List. 2012. "Digging into background risk:

experiments with farmers and students." American Journal of Agricultural

Economics 94 (2):457-463.

3.9 References

114

Holden, Stein T., and John Quiggin. 2016. "Climate risk and state-contingent

technology adoption: shocks, drought tolerance and preferences." European

Review of Agricultural Economics.



Just, Richard E., Linda Calvin, and John Quiggin. 1999. "Adverse selection in crop

insurance: Actuarial and asymmetric information incentives." American Journal of

Agricultural Economics 81 (4):834-849.








(1):37-48.

Menapace, Luisa, Gregory Colson, and Roberta Raffaelli. 2013. "Risk aversion,

subjective beliefs, and farmer risk management strategies." American Journal of


Menapace, Luisa, Gregory Colson, and Roberta Raffaelli. 2015. "A comparison of





42:767-780.




Meyer, Jack. 2002. "Expected Utility as a Paradigm for Decision Making in

Agriculture." In A Comprehensive Assessment of the Role of Risk in U.S.


Springer US.

3.9 References

115

Mishra, Ashok K., and Hisham S. El-Osta. 2002. "Managing risk in agriculture through

hedging and crop insurance: what does a national survey reveal?" Agricultural

Finance Review 62 (2):135-148.

Musser, Wesley N., and George F. Patrick. 2002. "How Much does Risk Really Matter

to Farmers?" In A Comprehensive Assessment of the Role of Risk in U.S.


Springer US.

Potter, Clive, and Matt Lobley. 1996. "The farm family life cycle, succession paths and

environmental change in Britain's countryside." Journal of Agricultural Economics

47 (1-4):172-190.

Reynaud, Arnaud, and Stéphane Couture. 2012. "Stability of risk preference measures:

results from a field experiment on French farmers." Theory and Decision 73

(2):203–221.

Savage, Leonard J. 1972. The Foundations of Statistics. New York USA: Courier

Corporation.

Schaper, Christian, Hauke Bronsema, and Ludwig Theuvsen. 2012. Risikomanagement

in der Landwirtschaft. Schriftenreihe des Landesamtes für Umwelt, Landwirtschaft

und Geologie 36, 36.

Shapiro, B. I., and Wade B. Brorsen. 1988. "Factors Affecting Farmers' Hedging

Decisions." North Central Journal of Agricultural Economics 10 (2):145-153.



van der Ploeg, Jan Douwe, and Dirk Roep. 2003. "Multifunctionality and rural

development: the actual situation in Europe." In Multifunctional agriculture: a new

paradigm for European agriculture and rural development, edited by Guido van

Huylenbroeck and Guy Durand, 37-53. Hampshire UK: Ashgate.


(1):1–24.





Velandia, Margarita, Roderick M. Rejesus, Thomas O. Knight, and Bruce J. Sherrick.

2009. "Factors Affecting Farmers' Utilization of Agricultural Risk Management

3.9 References

116

Tools: The Case of Crop Insurance, Forward Contracting, and Spreading Sales."

Journal of Agricultural and Applied Economics 41 (01):107-123.

Verbeek, Marno. 2008. A guide to modern econometrics. Vol. 4. New York USA: John

Wiley & Sons.

Weber, Elke U., Ann‐Renee Blais, and Nancy E. Betz. 2002. "A domain‐specific risk‐attitude scale: Measuring risk perceptions and risk behaviors." Journal of

Behavioral Decision Making 15 (4):263–290.

Weesie, Jeroen. 1999. "Seemingly unrelated estimation and the cluster-adjusted

sandwich estimator." Stata Technical Bulletin 52:34-47.

Zikmund-Fisher, Brian J., Dylan M. Smith, Peter A. Ubel, and Angela Fagerlin. 2007.

"Validation of the Subjective Numeracy Scale: Effects of Low Numeracy on

Comprehension of Risk Communications and Utility Elicitations." Medical

Decision Making 27 (5):663-671.

3.10 Appendix 3.A

117

3.10 Appendix 3.A

Sources of risk

Table 3.A.1: Sources of risk included in the survey to elicit overall risk perception scores

Sources of risk

Market- and price risks

Price volatility on sales market

Price volatility on procurement market

Increasing tenure prices

Increasing fodder prices

Weakening of producers due to increasing purchaser power

Institutional risks

Further decreasing of EU direct payments

Tightening of cross compliance

Increasing regulations for animal breeding (e.g. animal welfare regulations)

Increasing regulations for crop production (e.g. environmental protection regulations)

Reduction of EU market supporting measures (e.g. tariffs)

Further greening of the agricultural policy

Limitations to agricultural construction law

Disappearance of markets

Production risks

Yield volatility due to climate change

Yield loss due to climatic extreme events (e.g. flood, hail)

Epidemic animal diseases

Difficulties to fight pests and diseases (resistances)

Reduced land availability

Financial risks

Liquidity shortage

Decreasing creditworthiness

Other risks

Limited availability of qualified workforce

Default of the main farm operator

Problems with meeting quality standards

Acceptance problems of livestock farming (e.g. protests against new built stables)

Acceptance problems of arable farming (e.g. monoculture in agriculture)

Contextualized MPL

Instructions (translated from German):

To make sure that you understand the payout structure regarding your participation in

this survey, please read the following instructions carefully:

Below you see a table including 10 different decision scenarios (rows) for possible

investment outcomes (A and B). Each row of the decision table contains a pair of

3.10 Appendix 3.A

118

choices between Option A and Option B. 10 out of 100 participants will be chosen

randomly as winners. If you are one of them your payout will be calculated as

following:

1) One of the rows is selected at random, and the Option (A or B) that you chose

in that row will be used to determine your earnings.

Example: We assume, row 1 was selected randomly and your selected choice is

investment A.

2) After one of the decisions has been randomly selected, another random number

is chosen to elicit the probability of your payout. This random number determines your

earnings for the Option (A or B) that you previously selected for the decision being

used.

Example: In row 1 we randomly make a selection out of 10 balls (1 green and 9 blue)

to determine your payout. If a blue ball is selected, the amount is 80.000€. The actual

payout is divided by 1.000. Thus, you receive a payout of 80€.

Figure 3.A.1: Example visual presentation MPL

3.10 Appendix 3.A

119

Risk management strategies

Table 3.A.2: Risk management strategies included in the survey

On-farm agriculture N On-farm non-agriculture N Off-farm N

Risk adapted production

(e.g. prudent choice of

sowing time)

34

Non-agricultural

diversification (e.g. direct

sales, tourism, bio energy

production)

29 Off-farm work 11

Use of robust or resistant

varieties and breeds 47 Holding liquidity reserves 42 Off-farm investments 24

Agricultural

diversification (e.g. mixed

agriculture)

27 Work harder or cut private

expenses 14 Yield insurance 35

Investment in new

technologies, that adjusts

my production to the

weather (e.g. irrigation)

7 Cooperation with other

farmers 35 Hail insurance 36

Multiple risk insurance

(e.g. hail + storm + heavy

rain + heavy frost)

5

Business liability

insurance + floor-

coverage + environmental

liability

41

Legal protection insurance 43

Trading on the commodity

futures exchange 4

Table 3.A.3: Pearson correlation coefficients of risk preferences elicited with different methods

N = 6133 MPL SA Ø BS BSProd BSMark BSFin BSAg

MPL 1.000

SA 0.649*** 1.000

Ø BS 0.402*** 0.688*** 1.000

BSProd 0.481*** 0.686*** 0.883*** 1.000

BSMark 0.335*** 0.523*** 0.845*** 0.738*** 1.000

BSFin 0.253** 0.536*** 0.782*** 0.531*** 0.455*** 1.000

BSAg 0.311** 0.605*** 0.914*** 0.768*** 0.739*** 0.634*** 1.000

** Significant at the 5% level, *** significant at the 1% level

33

Note that we excluded three non-responding farmers for this analysis (two missing for the

MPL, one missing for the SA).

3.10 Appendix 3.A

120

Figure 3.A.2: Kernel density plots of three different risk preference elicitation methods

3.10 Appendix 3.A

121

Table 3.A.4: Spearman correlation coefficients of risk preferences elicited with different methods and risk perception in different domains

N = 6034 MPL SA Ø BS BSProd BSMark BSFin BSAg PerMarkRisk PerPolRisks PerProdRisk PerFinRisk PerOtherRisk

MPL 1.000

SA 0.724*** 1.000

Ø BS 0.456*** 0.674*** 1.000

BSProd 0.500*** 0.679*** 0.850*** 1.000

BSMark 0.391*** 0.500*** 0.799*** 0.682*** 1.000

BSFin 0.315** 0.540*** 0.744*** 0.499*** 0.411*** 1.000

BSAg 0.384*** 0.599*** 0.903*** 0.747*** 0.695*** 0.588*** 1.000

PerMarkRisk 0.083 0.089 0.063 0.078 0.014 0.001 0.051 1.000

PerPolRisks 0.173 0.126 0.142 0.123 0.045 0.066 0.090 0.539*** 1.000

PerProdRisk 0.079 0.026 -0.013 -0.044 -0.004 0.007 -0.067 0.379*** 0.483*** 1.000

PerFinRisk -0.107 -0.217* -0.124 -0.078 -0.124 -0.241* -0.104 0.467*** 0.251* 0.387*** 1.000

PerOtherRisk -0.090 -0.093 -0.075 -0.107 -0.087 -0.056 -0.125 0.411** 0.543*** 0.472*** 0.321** 1.000

34

Note that we have excluded four non-responding farmers for this analysis (two missing for the MPL, one missing for the SA, one missing

for the risk perception).

3.10 Appendix 3.A

122

Table 3.A.5: Multinomial logit estimates for MPL, SA and average BS

MPL SA Ø BS

N = 56 On-farm agriculture On-farm non-

agriculture On-farm agriculture

On-farm non-

agriculture On-farm agriculture

On-farm non-

agriculture

Farmer

characteristics

Risk aversion 2.564* (1.314) 0.304 (0.818) -0.046 (0.422) 0.448* (0.229) 1.233 (0.982) 1.712* (0.855)

Age -1.503* (0.804) 0.737 (0.520) -1.259 (0.769) 0.723 (0.577) -1.579* (0.875) 0.156 (0.637)

Age^2 0.018** (0.009) -0.008 (0.006) 0.015* (0.009) -0.008 (0.006) 0.019* (0.010) -0.002 (0.007)

Educ 0.611 (0.528) -0.415 (0.415) 1.280 (0.812) -0.523 (0.511) 1.072* (0.647) -0.763 (0.563)

Optim -0.411 (0.874) -0.530 (0.631) -0.109 (0.908) -0.434 (0.487) 0.469 (0.802) 0.157 (0.536)

SN 4.881*** (1.855) 1.821 (1.121) 4.382** (1.770) 1.724* (0.940) 4.446*** (1.452) 1.539* (0.802)

AgricTrain -0.122 (0.585) 0.408 (0.478) -0.391 (0.705) 0.259 (0.437) -0.573 (0.641) 0.205 (0.458)

PerMarkRisk -1.397** (0.556) -0.238 (0.176) -1.386** (0.543) -0.309* (0.174) -1.526*** (0.570) -0.434* (0.237)

PerPolRisks 0.158 (0.316) 0.138 (0.188) 0.319 (0.301) 0.125 (0.198) 0.266 (0.284) 0.157 (0.219)

PerProdRisk -0.186 (0.244) -0.089 (0.173) -0.133 (0.253) -0.123 (0.170) -0.186 (0.263) -0.103 (0.196)

PerFinRisk -0.184 (0.187) 0.052 (0.128) -0.321 (0.263) 0.130 (0.140) -0.177 (0.241) 0.193 (0.189)

PerOtherRisk 0.558 (0.420) 0.009 (0.252) 0.486 (0.375) -0.017 (0.239) 0.588 (0.435) -0.017 (0.249)

Optim 13.802*** (4.605) 3.159** (1.414) 12.571*** (4.070) 2.844** (1.139) 12.555*** (3.388) 3.193** (1.259)

Household

characteristics

WF 0.097 (0.763) 0.396 (0.547) 0.166 (0.880) 0.174 (0.575) 0.332 (0.713) 0.569 (0.466)

FarmSuc -1.052 (2.176) 2.801 (1.782) -1.871 (2.637) 2.824* (1.618) -1.481 (1.851) 2.483** (1.184)

HHS -0.582 (0.549) 0.249 (0.372) -0.418 (0.488) 0.197 (0.404) -0.401 (0.519) 0.262 (0.439)

Farm

characteristics

AgricArea -0.031* (0.018) 0.003 (0.006) -0.028 (0.020) 0.003 (0.006) -0.032* (0.019) 0.002 (0.005)

PropRentLand 2.164 (4.179) -7.298* (2.683) 2.136 (3.845) -6.471*** (2.323) 1.578 (3.968) -7.857*** (2.818)

LivePig 0.001 (0.001) 0.001 (0.000) 0.000 (0.001) 0.001* (0.000) 0.001 (0.001) 0.001** (0.001)

LiveCattle 0.014 (0.008) 0.001 (0.005) 0.010 (0.007) 0.004 (0.004) 0.013* (0.007) 0.006 (0.004)

Constant 18.177 (17.544) -19.537 (12.547) 9.972 (15.241) -18.306 (14.544) 14.856 (16.923) -7.730 (14.013)

Wald Chi2(40) 74.3 73.56 86.98

Log

pseudolikelihood -33.895995 -34.297416 -34.121888

Prob > chi2 = 0.0013 0.0010 0.0000

3.10 Appendix 3.A

123

Table 3.A.6: Multinomial logit estimates for BS production, BS marketing and prices, BS finances and BS agriculture generally

BSProd BSMark BSFin BSAg

N = 56 On-farm

agriculture

On-farm non-

agriculture

On-farm

agriculture

On-farm non-

agriculture

On-farm

agriculture

On-farm non-

agriculture

On-farm

agriculture

On-farm non-

agriculture

Farmer

characteristics

Risk aversion 0.741 (0.847) 0.766 (0.546) 0.502 (0.720) 1.365* (0.721) 1.532** (0.731) 0.816 (0.675) -0.615 (0.748) 0.900 (0.607)

Age -1.422* (0.849) 0.459 (0.503) -1.340* (0.766) 0.259 (0.502) -1.784** (0.731) 0.363 (0.669) -1.592* (0.888) 0.216 (0.648)

Age^2 0.017* (0.010) -0.005 (0.005) 0.016* (0.009) -0.003 (0.005) 0.021** (0.008) -0.004 (0.007) 0.019* (0.010) -0.002 (0.007)

Educ 1.260** (0.632) -0.393 (0.417) 1.097 (0.719) -0.696 (0.479) 1.020 (0.743) -0.597 (0.649) 1.459** (0.645) -0.473 (0.401)

Optim 0.238 (0.771) -0.252 (0.530) 0.208 (0.928) -0.055 (0.517) 1.118 (0.696) 0.084 (0.720) -0.125 (1.255) -0.157 (0.524)

SN 4.046*** (1.531) 1.483* (0.865) 4.350*** (1.626) 1.848** (0.745) 4.234*** (1.625) 1.676 (1.036) 4.916*** (1.548) 1.375** (0.724)

AgricTrain -0.589 (0.647) 0.221 (0.458) -0.546 (0.608) -0.099 (0.464) -0.935 (0.623) 0.246 (0.502) -0.072 (0.694) 0.295 (0.420)

PerMarkRisk -1.334*** (0.475) -0.284* (0.165) -1.457** (0.642) -0.367* (0.222) -1.561** (0.449) -0.431** (0.207) -1.434 (0.535) -0.208 (0.173)

PerPolRisks 0.188 (0.280) 0.103 (0.195) 0.307 (0.319) 0.120 (0.230) 0.347 (0.250) 0.274 (0.192) 0.372 (0.273) 0.098 (0.188)

PerProdRisk -0.154 (0.246) -0.064 (0.168) -0.101 (0.237) 0.000 (0.219) -0.230 (0.282) -0.145 (0.209) -0.124 (0.237) -0.043 (0.187)

PerFinRisk -0.262 (0.243) 0.080 (0.138) -0.250 (0.248) 0.091 (0.137) -0.126 (0.251) 0.171 (0.202) -0.317 (0.249) 0.076 (0.126)

PerOtherRisk 0.620 (0.433) 0.037 (0.231) 0.537 (0.434) -0.028 (0.265) 0.468 (0.396) -0.129 (0.262) 0.397 (0.464) -0.090 (0.229)

Optim 11.862***

(3.137) 2.687** (1.112)

12.493***

(4.161) 2.992** (1.179)

12.154***

(3.323) 2.918** (1.262)

14.086***

(3.587) 2.856*** (1.167)

Household

characteristics

WF 0.132 (0.724) 0.412 (0.515) 0.199 (0.817) 0.506 (0.438) -0.032 (0.802) 0.474 (0.573) -0.070 (0.868) 0.335 (0.466)

FarmSuc -1.340 (1.884) 2.661* (1.503) -1.594 (2.012) 2.293** (1.143) -0.933 (1.617) 2.428 (1.650) -2.627 (2.089) 2.037 (1.293)

HHS -0.357 (0.500) 0.238 (0.396) -0.453 (0.493) 0.117 (0.356) -0.482 (0.553) 0.274 (0.494) -0.352 (0.474) 0.230 (0.435)

Farm

characteristics

AgricArea -0.030 (0.021) 0.001 (0.005) -0.027 (0.017) 0.001 (0.005) -0.027 (0.019) 0.005 (0.006) -0.025 (0.018) 0.003 (0.004)

PropRentLand 2.916 (4.060) -6.034*** (2.206) 1.568 (4.032) -7.236*** (2.478) 2.295 (3.725) -7.374** (2.903) 1.404 (3.762) -6.865*** (2.437)

LivePig 0.001 (0.001) 0.001** (0.000) 0.000 (0.001) 0.001** (0.000) 0.001 (0.002) 0.001* (0.001) 0.000 (0.001) 0.001** (0.000)

LiveCattle 0.012 (0.007) 0.003 (0.004) 0.010 (0.007) 0.007 (0.004) 0.017** (0.007) 0.003 (0.004) 0.009 (0.008) 0.005 (0.004)

Constant 11.118 (15.486) -14.949 (12.029) 11.691 (15.813) -8.771 (11.801) 21.273 (13.913) -9.818 (16.414) 14.824 (17.644) -8.790 (12.858)

Wald Chi2(40) 106.14 86.47 96.38 104.62

Log

pseudolikelihood -35.257268 -33.980495 -33.628467 -33.759636

Prob > chi2 = 0.0000 0.0000 0.0000 0.0000

124

Chapter 4

Determinants and motives for agritourism activities: A

German case study*

* Meraner, M., K. Holtmann and R. Finger (under review)

125

Abstract

Farm diversification is of key relevance for the agricultural sector and for

rural development. We investigate the determinants and motives for

agritourism activities. More specifically, we focus on four aspects, namely i)

motives for agritourism diversification, ii) the role of farmwomen in the

agritourism venture, iii) farm and farmers’ characteristics and iv) farmers’

risk preferences. Our analysis is based on 33 interviews conducted with

agritourism and non-agritourism farmers located within and in the

surroundings of the city of Muenster, Germany. Our results show that the

uptake of agritourism is mainly driven by pull motives (e.g. taking advantage

of market opportunities or joy of working in agritourism). Focusing on the

role of farmwomen, we find that even though decisions are often made

jointly, farmwomen are more often responsible for the realization of

activities than the initialization and planning. Comparing agritourism and

non-agritourism farms, we find the former to be smaller, more diversified

and more often run by female farm operators. Furthermore, we find

agritourism farmers to be more risk averse. Agritourism creates a source of

income that is independent from unstable agricultural income and thus

contributes to smooth volatile agricultural income.

Keywords: Farm diversification, agritourism, motives, decision

making, risk preferences

JEL classification: Q12, Z30

4.1 Introduction

126

4.1 Introduction

Farm diversification is a key component of farm-survival and rural development in

many countries of the world and allows farmers to stabilize or increase their income

(e.g. Barrett, Reardon, and Webb, 2001, McNamara and Weiss, 2005). For instance,

one out of three farmers in the European Union has diversified into non-agricultural

activities (European Parliamentary Research Service, 2016), which also represents the

promotion of farm-diversification in the European Union since the 1980s with

subsidies and support for agricultural diversification being established (Ilbery and

Bowler, 1998). Among various diversification measures, agritourism is of central

relevance as it also contributes to rural development and improves economic

opportunities and accessibility in disadvantaged rural regions (European Commission,

1990). Farm diversification in the broad sense and agritourism in the narrow implies

that farms reallocate resources (e.g. capital or labor) away from the production of

conventional crops and livestock to generate income (McInerney and Turner, 1991,

Ilbery, 1991). Our analysis focusses on farmers’ motives to take up agritourism

activities, which we define as activities that incorporate both a working farm

environment and a commercial tourism component (Weaver and Fennell, 1997).

Four streams of literature have investigated farmers’ decisions to engage in

agritourism. First, there is a focus on stated motives of farm managers behind farm

diversification in order to understand all determinants leading to agritourism (Haugen

and Vik, 2008, McGehee and Kim, 2004, Ollenburg and Buckley, 2007). Second, there

4.1 Introduction

127

is a large body of literature focusing on finding observable farm, farmer and household

characteristics determining the agritourism decision (Meraner et al., 2015, McNally,

2001, Ilbery et al., 1998, McNamara and Weiss, 2005, Mishra, El‐Osta, and Sandretto,

2004, Bagi and Reeder, 2012). Third, some studies focus on agricultural diversification

including agritourism as a farm risk management strategy used to smooth volatile

agricultural income (Mishra, El‐Osta, and Sandretto, 2004, Meuwissen, Huirne, and

Hardaker, 2001, Kostov and Lingard, 2003). This literature builds on the hypothesis

that farmer’s choice to engage in agritourism activities is driven by individual risk

preferences. Fourth, many studies focus specifically on the role of farmwomen in the

agritourism business (Sharpley and Vass, 2006, Hansson et al., 2013, Hjalager, 1996,

McGehee, Kim, and Jennings, 2007). Hansson et al. (2013) emphasis on the role of

farmwomen in different stages of the agritourism business (planning and managing).

Furthermore, Haugen and Vik (2008) and Brandth and Haugen (2007) focus on the

gendered nature of different agritourism activities leading to differences in the intensity

of farmwomen’s participation in different agritourism activities.

Despite this rich set of literature, no study has combined all four aspects. More

specifically, no study has provided a coherent analysis of farmers’ motives, farm and

farmers’ characteristics and risk preferences driving agritourism decisions. We

contribute filling gaps in the literature by combining the four dimensions in our

analysis. To this end, we conduct interviews with German farmers to understand

farmers’ motives influencing the agritourism decision. Moreover, we obtain farm,

farmer and household characteristics and elicit farmers’ risk preferences using

4.2 Theoretical framework

128

experimental risk preference elicitation methods. For the latter, farmers with and

without agritourism are compared with each other to identify determinants

characterizing farms diversifying in agritourism. Furthermore, we analyze the specific

role of farmwomen in different stages of the agritourism business for a variety of

different agritourism activities. Against this background, this study focuses on

answering the following research questions: i) Which motives underlie the decision to

engage in agritourism? ii) To what extend are farmwomen participating in the

initialization and implementation process of agritourism activities? iii) Which, if any,

farm, farmer and household characteristics including risk preferences are deterministic

of agritourism farms? Findings to these research questions can support the

development of more tailored extension services and better tailored policies aiming to

support farm diversification into agritourism.

The remainder of this paper is organized as follows. The subsequent section

presents the determinants of agritourism. This is followed by a description of the

methodology of data collection and data. Subsequently, we present the results based on

our research questions and a discussion of our analysis.


In this section, we discuss the theoretical frameworks as well as the key literature

underlying the four blocks of relevance for our analysis, i.e. i) motives for agritourism

diversification, ii) the role of farmwomen in the agritourism venture, iii) farm and


129

farmers’ characteristics and iv) farmers’ risk preferences. Based on these theoretical

frameworks, aspects for the operationalization in our interviews are derived.

4.2.1 Motives for agritourism initiation

Defining the establishment of a new venture outside traditional agriculture as an

entrepreneurial activity (McElwee, 2008, Segal, Borgia, and Schoenfeld, 2005),

farmers are either pushed or pulled into the startup of new ventures. In order to group

the various motives behind the farmers’ decision for agritourism entrepreneurship

literature suggests the distinction between “opportunity driven” (pull factor) or

“necessity driven” (push factor) decisions. Thus, push factors comprise a

dissatisfaction with the current situation and agricultural business environment, where

the farmer is pushed to activities outside agricultural production by the necessity to

seek extra income, become self-employed or decrease agricultural risks (Busby and

Rendle, 2000, Hansson et al., 2013). In contrast, pull factors include motives based on

a favorable economic environment outside traditional agriculture. This includes the

existence of emerging local demand or markets, or excess labor capacities.

In order to describe motives in a more refined way, further sub-categories are

often used (e.g. Nickerson, Black, and McCool, 2001, McGehee and Kim, 2004,

Barbieri, 2009, Ollenburg and Buckley, 2007, Di Domenico and Miller, 2012). Based

on this literature we classify the set of motives included in our study into four

categories, namely: i) farm-survival, ii) intrinsic, iii) extrinsic, and iv) family motives.

Farm-survival motives are mainly necessity driven. Some studies find that the need of

extra income and the insufficient income from agriculture are the main motives of


130

farmers engaging in agritourism (McGehee and Kim, 2004, Barbieri, 2009).

Furthermore, Nickerson, Black, and McCool (2001) and McGehee and Kim (2004)

show that optimal usage of farm resources and minimizing the fluctuations in

agricultural income are important motives. Hansson et al. (2013) and Barbieri (2009)

additionally include the farmers’ wish to reduce debts in the business. Intrinsic motives

are concerned with the farmers’ lifestyle. Vik and McElwee (2011) find that the wish

to create something i.e. learn and acquire new skills is an important motive for

agritourism farmers. Barbieri (2009) and Di Domenico and Miller (2012) find that the

wish to continue farming and enhancement of the personal and family quality of life to

be important motives for farmers. Furthermore, joy in the work, passion to work with

people and turning a hobby into a career are motives included in this category (Vik and

McElwee, 2011, Ollenburg and Buckley, 2007). Medhurst and Segrave (2007) claim

that independence, self-sufficiency, flexibility and to work at home are important

intrinsic motives. Extrinsic motives are those focusing on the external demand i.e.

opportunity based factors, determined externally (e.g. by the geographical, political, or

demand environment). Barbieri (2009) and Hansson et al. (2013) find evidence that

motivation for agritourism farmers is driven by market needs and growth opportunities

outside traditional agriculture (possibly stimulated by the farms location). Furthermore,

Nickerson, Black, and McCool (2001) claim that farmers can be motivated by

governmental incentives (e.g. taxes or subsidies). Hansson et al. (2013) also considers

the wish to gain independence from agricultural policy. Family motives include those

motives concerned with the farm family. Hansson et al. (2013) and Barbieri (2009)


131

highlight the importance of employment for family members (and opportunities to take

better care of the children and household). Furthermore, Ollenburg and Buckley (2007)

and Hansson et al. (2013) include the need to keep the business in family ownership

and create an employment opportunity for future generations.

4.2.2 Role of farmwomen in the agritourism venture

Agritourism ventures demand flexibility regarding time management between the

tourism businesses and farming business thus, agritourism is largely family based,

requiring the involvement of the farming couple. In this light Phelan and Sharpley

(2011) propose an expansion of the focus on the main farm in entrepreneurial theory to

the farmers spouse (i.e. copreneurship). Recent studies in Europe find that the main

workload of the agritourism business lies with the female partners of the farm family

(Sharpley and Vass, 2006, Nilsson, 2002, Haugen and Vik, 2008, Busby and Rendle,

2000). In an early study Hjalager (1996) argues that agritourism can be a way of

making the work of female farm family members profitable as many agritourism

activities find their origins among labor traditionally performed by women on the farm.

Moreover, Medhurst and Segrave (2007) and McGehee, Kim, and Jennings (2007)

underline the importance of agritourism for the women’s financial independence. Yet,

Haugen and Vik (2008) argue that the intensity of the involvement of women in

tourism activities follows traditional lines, i.e. they find that women are more engaged

in accommodation and food-serving than in adventure activities. Furthermore, Hansson

et al. (2013) suggest that the spouse’s involvement in agritourism activities changes in

the different stages of the creation of the new venture (i.e. initialization/planning and


132

realization/managing). Next to investigating for the relevance of female farm managers

for the uptake of agritourism activities, we explicitly obtain information on the

involvement of farmwomen in the initialization and the realization of agritourism

activities at the farm.

4.2.3 Farm, farmer and household characteristics

The extent to which resources are allocated to the nonagricultural income activities is

influenced by the farms given external decision making environment as well as internal

factors as discussed above (Ilbery, 1991, Evans and Ilbery, 1989).

On the farm business level Haugen and Vik (2008) argue that agritourism is an

aspect of an economic survival strategy, ensuring farm survival without the loss of

independence trough taking off-farm employment. Thus, we expect that agritourism

farms are operated more likely as full time activity. Barbieri (2009) finds that

agritourism farms are more likely to have a portfolio of on-farm diversification

activities as there are synergistic relationships between agritourism and other activities.

The type of farming is found to be closely linked to the diversification

decision. McNally (2001) found that the seasonality of the farming activity influences

the diversification decision. Specifically, Meraner et al. (2015) conclude that livestock

farms with a constantly high labor demand are less likely to have spare time to develop

a diversification strategy off-season.

A very common finding in literature is that farm size has a significant effect on

the diversification decision. However, the definition and measurement unit of farm size

inconsistent in the literature. McNally (2001) and McNamara and Weiss (2005)


133

suggest that larger farms in terms of average net income and number of livestock units

are more likely to be diversified, since they can allocate and exploit available resources

more efficiently. Contrarily, Vik and McElwee (2011) and Meraner et al. (2015) find

that larger agricultural size in terms of hectares of land, decreases the farms’

probability to engage in agritourism, larger farms are more likely to use possible

economies of scale and choose a specialization strategy.

The farmers’ age has been found to influence the diversification decision.

Barbieri (2009) and Meraner et al. (2015) argue that younger farmers have stronger

needs to create long-term ties and the need to strengthen the farm business, thus they

are more likely to diversify the business. However, Haugen and Vik (2008) find that

farmers with agritourism do not start the business at a very young age when financial

resources are scarce. Benjamin and Kimhi (2006) find in this context that farm

diversification is more attractive for not too young farmers, with a peak for middle-

aged farmers. This peak phase is followed by a period of increased income needs, due

to family building, with a reduced diversification probability. In a later stage (when the

children have left the house), income needs are decreasing and farm diversification is

attractive again.

Furthermore, Barbieri (2009) and Haugen and Vik (2008) find that agritourism

farmers are more likely to have a higher general education, as well as higher degrees of

agricultural education. They argue that higher agricultural education goes alongside

with stronger occupational identity with farming as success factor for a profitable

business within agritourism (Sharpley and Vass, 2006, Di Domenico and Miller,


134

2012). Thus, we additionally include the emotional attachment to agriculture in our

analysis.

An important factor influencing the agritourism decision on the household

level is the availability of family workforce. Mishra, El‐Osta, and Sandretto (2004)

suggest that larger families have a stronger need to create employment opportunities on

the farm, leading to more agricultural diversification. Barbieri (2009) finds that

households with adult children at home are more likely to engage in agritourism, thus

we include whether the succession of the farm business is secure or not. Furthermore,

Haugen and Vik (2008) argue that agritourism is a family based operation and find that

married or cohabiting farm couples are more likely to engage in agritourism. A full list

of all variables included in our analysis based on the presented literature is found in

Table 4.3.

4.2.4 Farmers’ risk preferences

Finally, farmers’ risk preferences are expected to influence decisions towards

agritourism. Farm diversification and agritourism create sources of income that are

independent from unstable agricultural income and thus contribute to smooth volatile

agricultural income (Mishra, El‐Osta, and Sandretto, 2004, Meuwissen, Huirne, and

Hardaker, 2001, Kostov and Lingard, 2003). Thus, risk averse farmers are assumed to

shift more of their resources away from agricultural production to diversify towards

activities such as agritourism in order to reach a reduction of the overall riskiness of

household income (McNamara and Weiss, 2005, Hardaker et al., 2004). To elicit

farmers’ risk preferences in surveys and interviews, a wide set of methodological

4.3 Sampling and data

135

approaches is available. For instance, there is a growing body of literature on the

farmers’ decision making focusing on the role of individual risk preferences elicited

using experimental risk preference elicitation methods (Meuwissen, Huirne, and

Hardaker, 2001, van Winsen et al., 2014, Menapace, Colson, and Raffaelli, 2013). In

addition, various self-assessment statements have been used in the literature to elicit

farmers’ risk preferences. With no method dominating the others and the observation

that different methods might even result in contrasting results (Hellerstein, Higgins,

and Horowitz, 2013, Reynaud and Couture, 2012, Crosetto and Filippin, 2015,

Menapace, Colson, and Raffaelli, 2015), we opted to include three risk preference

elicitation methods and compare risk aversion of agritourism farmers and non-

agritourism farmers. More specifically, we use a lottery based on a multiple price list

following Holt and Laury (2002), a self-assessment of risk preferences (Menapace,

Colson, and Raffaelli, 2015) and five agricultural business statements adapted from

Meuwissen, Huirne, and Hardaker (2001). In order to reduce the potential to obtain

contrasting results, contextualized elicitation methods are used.


4.3.1 Sampling and case study area

Agritourism has a long tradition in Germany, latest documents on farm tourism go as

far as 150 years back (Nilsson, 2002). Most of the existing German research in the

field is focusing on the costal and alpine regions with a high density of agritourism

farms (Oppermann, 1996, 1997, Lehner-Hilmer, 1999). Next to natural areas and


136

attractive landscapes, however, also the proximity to urban areas was found to be a

contributing factor to observe agritourism activities (Lange et al., 2013, Ilbery, 1991,

Le Grand and van Meekeren, 2008). Market proximity is associated with increased

demand and marketing opportunities which are found to stimulate peri-urban farmers

to identify market niches, innovate and adapt to new demands. Thus, we focus in our

analysis on the urban area of the city of Muenster in the state of North-Rhine-

Westphalia, which is characterized by a combination of attractive landscapes and large

urban areas36

(Figure 4.1). By including farms in a relatively small area, we control for

external factors influencing the diversification decision. Thus, we can assume that

factors like infrastructure, soil quality, market opportunities as well as attractiveness of

the landscape are similar for all farms in the area (Busby and Rendle, 2000, Pfeifer et

al., 2009, Ilbery, 1991, Lange et al., 2013, Walford, 2001).

We used the online platform of the Chamber of Agriculture “Landservice”

(http://www.landservice.de) to obtain a list of farms being active in agritourism, which

we complemented with internet searches using the activities as key words. As defined

above we classify all farms with a commercial tourism component as agritourism

farms, this includes accommodation, organizing events, leisure, sport or recreational

activities, gastronomy, renting out locations for events and conferences and equestrian

businesses (e.g. horseback riding, pension horses). The choice of interview partners

followed purposive and snowball approaches. We identified in total 26 farmers

36

More specifically, we focus on the city of Muenster as well as municipalities within a radius

of 25km around the city center comprising the city of Muenster and its eastern surrounding

municipalities (Telgte, Everswinkel, Sendenhorst and Drensteinfurt) (Figure 4.1). The dark

grey fields in the left map of Figure 4.1 depict the city center of Muenster, whereas the

surrounding areas are lighter.


137

involved with agritourism in the selected area. We retrieved a sample of 17 farms with

agritourism activities (four farmers did not respond to the contacting approaches, three

where not willing to participate and two had no agricultural activity anymore).

Additionally, 16 farmers without agritourism activity where interviewed. Leading to a

total sample of 33 interviews and a response rate of 79%. The farm location of

interviewed farmers is depicted in Figure 4.1.

The potential interview partners where contacted via phone and e-mail to

schedule appointments for on-site computer assisted face-to-face interviews in January

2016. The farms without agritourism activities where selected using snowball

sampling. Each interviewed agritourism farmer was asked at the end of the interview to

give information on five neighbors who could potentially participate in the study. Out

of the five potential future interview partners we selected randomly two, to contact and

schedule appointments. This approaches’ biggest advantage is that we could interview

direct neighbors with seemingly similar external preconditions (i.e. distance to

potential markets, attractiveness of the landscape) but different choices of farm

management strategies. Additionally, snowball sampling increases the credibility of

the research, as participants are involved in the research process and it is cost effective.

Nevertheless, when using this sampling method the anonymity between participants

cannot be obtained, additionally there might be a bias as participants choose people

they know and share the same viewpoint (King and Horrocks, 2010). To interview a

large number of farmers we used structured, closed-ended interview questions.

Furthermore, the biggest advantage of structured interview questions is the increased


138

reliability of results (King and Horrocks, 2010). The interviews took on average two

hours. A pre-test was conducted with five farmers in the area. The survey consisted of

four parts including questions on: (1) the farms diversification activities, (2) motives

for agritourism, (3) the spouse’s involvement with agritourism and (4) the farmer

(including risk preferences), the farm household and the farm business.

Figure 4.1: Location of case study area

4.3.2 Data description

To analyze a holistic picture of all diversification activities the farmers engage in, we

presented a list of 14 on-farm agricultural diversification activities Namely we include

five agritourism activities: gastronomy, accommodation, equestrian business, renting

out locations, organizing events and ten other on-farm diversification activities: direct

marketing, processing of agricultural products, selling Christmas trees, social farming,


139

renewable wind energy, biogas production, photovoltaic systems, agricultural labor for

others and woodwork.

To identify the motives underlying farmers’ decision to engage in agritourism

activities we identified 24 motives that are included in our analysis based on the

literature presented in section 4.2.1 these motives are categorized in two ways. First,

we distinguish push and pull motives, which allows us to identify general drivers of

agritourism engagement. Second, we distinguish these motives in groups of farm-

survival, intrinsic, extrinsic and family motives (the complete list is presented in Table

4.1). To capture farmers’ motives, we adapted a measurement scale used in other

studies in the field (Hansson et al., 2013, Sharpley and Vass, 2006, Barbieri, 2009, Vik

and McElwee, 2011, Nickerson, Black, and McCool, 2001). Respondents were asked

to indicate on a Likert scale of 1-5 the degree of importance of each motive

determining the decision to engage in agritourism. In order to ensure that respondents

could distinguish between response options in a meaningful way, the anchors (1) very

important; (2) fairly important; (3) moderately important; (4) somewhat important; and

(5) not important were used. This scale was chosen based on pre-test results, it is

linked to German schooling grades (1 = “very good” and 5 = “deficient”). In order to

ease comparability with other research in the field we use inverted motive variables the

further analysis. Furthermore, the interviewee explained the scale verbally and

presented the full list of motives on paper so that participants could additionally read

the motives if necessary. No opt-out alternatives were provided. In Table 4.1 the full

list of motives included in the analysis is presented, to ease understanding.


140

Furthermore, farmers could also freely add in an open question motives they perceived

as important and where not included in the list provided.

The role of women in the agritourism business is assessed on the planning and

management level of the agritourism business. In particular, the involvement in the

initiation/planning and actual realization/management of farmwomen in the

agritourism venture was measured on a 5 point Likert scale, adopted from Hansson et

al. (2013). More specifically, respondents were presented with the following options:

1. My spouse was solely responsible for initiation/planning of the agritourism venture.

2. My spouse took more part than me in the initiation/planning of the agritourism

venture. 3. My spouse and I took equal parts in the initiation/planning of the

agritourism venture. 4. My spouse took less part than me in the initiation/planning of

the agritourism venture. 5. My spouse did not take part at all in the initiation/planning

of the agritourism venture. Accordingly, respondents were asked to indicate their

spouses’ role in the realization/management of the agritourism venture on a 5 point

Likert scale. Keeping in line with other research in the filed the scale was inverted for

further analysis so that higher values indicate more involvement of farmwomen. In

three cases where the interview partner was female we adapted the responses to capture

the role of farmwomen.

The farm, farmer and household characteristics included in this study to

compare agritourism and non-agritourism farmers are explained in detail in Table 4.3.

We assess risk preferences by using three different methods, namely i) a standard

multiple price list (MPL) as introduced by Holt and Laury (2002), ii) a self-assessment


141

of risk preferences (Menapace, Colson, and Raffaelli, 2015) and, iii) five agricultural

business statements adapted from Meuwissen, Huirne, and Hardaker (2001).

The MPL is adopted from Holt and Laury (2002) including a list of ten rows

with ten unique choices between a safe option (A) and a risky option (B). The number

of safe choices is determining the farmers risk preferences ranging from 1 = “very risk

averse” to 9 = “very risk loving” (the last row includes a control question not

considered in the further analysis of risk preferences). We incentivized the MPL using

a realistic payout structure ranging from 5€ to 192.5€, with payouts being made to 10%

of the participating farmers (see Maart-Noelck and Musshoff, 2013). Furthermore, we

frame the MPL in an agricultural setting to assimilate the real world decision-making

context of farmers. An example for the lottery is presented in the Appendix 4.A. For

the self-assessment of risk preferences we ask respondents to indicate their general

attitude towards risk on a scale from 0 to 10 (adapted from Menapace, Colson, and

Raffaelli, 2015): ‘How do you see yourself personally: are you generally willing to

take risks, or do you try to avoid them wherever possible? Please indicate your attitude

towards risk on a scale from 0 to 10 where 0 means “not at all willing to take risks”

and 10 “very willing to take risks”.’ Additionally, we include four business statements

measuring the farmers relative risk aversion in four risk dimensions relevant for

agriculture as proposed by Meuwissen, Huirne, and Hardaker (2001). Participants are

asked to indicate the extend of agreement (1 = “fully agree” to 5 = “don’t agree”) with

the following four statements: 1. I am willing to take more risks than my colleagues

with respect to production. 2. I am willing to take more risks than my colleagues with

4.4 Results

142

respect to marketing. 3. I am willing to take more risks than my colleagues with respect

to financial issues. 4. I am willing to take more risks than my colleagues with respect

to agriculture generally. In order to incentivize participation we offered all

participating farmers a feedback report including the study’s results. To analyze the

differences in characteristics of agritourism farms and non-agritourism farms we use

the Mann-Whitney-Wilcoxon Test. For the analysis and visualizations presented in the

article the statistical software R (packages: psych, ggplot2 and maps) is used (R Core

Team, 2016). All raw data, codes and the complete survey are available from the

authors upon request.

4.4 Results

4.4.1 Motives behind agritourism

Average scores and standard deviations of the measurement items used to capture

farmers’ motives for starting agritourism are shown in Table 4.1. The strongest motive

for all interviewed farmers is the desire to use new market opportunities, followed by

the joy to work in agritourism and a passion for working with people. Least important

motives are tax reasons and the wish to create employment opportunities for the future

generation.

4.4 Results

143

Table 4.1: Descriptive statistics of suggested motives for starting an agritourism activity

All agritourism

farms

(N = 17)

Total

rank

pull/push Mean SD

Farm-survival motives

Agritourism as a form of extra income push 4.06 1.14 5

Minimizing income risk stemming from

agriculture push 3.94 1.30 6

Better use of existing or idle capacities, such

as facilities, machinery, or area pull 3.94 1.14 7

Expansion of agricultural production

impossible push 3.24 1.15 14

Insufficient income from agriculture

(livelihood) push 3.18 1.38 15

Reduction of overall business debt push 1.65 1.00 21

Average of all farm-survival motives 3.33 0.42

Intrinsic motives

Joy to work in agritourism pull 4.29 0.92 2

Passion for working with people pull 3.71 0.92 3

Desire to pursue personal interests and realize

your own interests pull 3.12 1.54 9

Independence, self-sufficiency, flexibility pull 3.18 1.47 10

Desire to learn and acquire new skills pull 4.24 0.75 11

An opportunity to continue to operate the

business (desire to keep working as a farmer) push 3.69 0.53 12

Desire to work at home, merge workplace

with home pull 3.76 1.15 16

Turn your hobby into a career pull 3.59 1.37 18

Average of all intrinsic motives 4.29 0.92

Family motives

Putting free family workforce to use push 3.18 1.38 17

A way to keep the business in family

ownership push 3.00 1.27 19

Ability to work from home to take better care

of children and household pull 2.06 1.25 20

Create an employment opportunity for future

generations push 1.18 0.53 23

Average of all family motives 2.35 0.64

4.4 Results

144

Table 4. 1 cont.: Descriptive statistics of suggested motives for starting an agritourism activity

All agritourism

farms

(N = 17)

Total

rank

pull/push Mean SD

Extrinsic motives

Use market opportunities (e.g. access to new

markets, higher demand) pull 4.53 0.80 1

Location of farm is convenient for agritourism pull 4.12 1.27 4

Growth opportunities/potential for business

bigger aside from traditional agricultural

production

pull 3.82 1.29 8

Gaining independence from agricultural

policy (e.g. increase in cross-compliance

requirements, increase in requirements for

livestock farming and plant based production)

push 3.41 1.70 13

Governmental support and subsidies (support

for conversion/reutilization of farm buildings,

rural development programs and projects)

pull 1.59 1.12 22

Tax reasons pull 1.06 0.24 24

Average of all extrinsic motives 3.09 0.62

Average of all push motives 3.11 0.40

Average of all pull motives 3.33 0.42

Note: The statements were measured on a Likert scale ranging from 1 to 5 (not important - very

important). The minimum score of all suggested motives was 1 and the maximum score was 5.

To further illustrate the differences between motive categories Figure 4.2

shows boxplots of motive categories for agritourism farmers. The left panel shows

box-plots of the average scores of each farm for each of the four categories. The right

panel shows the average of motives for each farm if grouped into pull and push factors.

We find that family and extrinsic motives are less important than farm-survival and

intrinsic motives in our sample. Furthermore, we find on average the largest

discrepancies, measured in standard deviations, with respect to the importance of

intrinsic motives. Whereas the importance of farm-survival motives is more similar

4.4 Results

145

within our sample. Along these lines, we find that pull factors are more relevant for the

agritourism decision in our sample. Furthermore, farmers could also add motives they

did not find in the list but where relevant to them in the decision to start the

agritourism venture. Many of the comments point towards economic reasons including

the synergetic effect of the agritourism venture and existing direct sale on the farm.

The statements comprised: i) “With the gastronomy (café) we hoped to be able to

advertise the on farm blueberry sale to a larger customer base.”, ii) “We organize

events, and public tours on the farm to attract more customers, and educate the

public.”, iii) “We wanted to attract people to visit the farm and be able to sell the

Christmas trees at higher prices.”

Figure 4.2: Differences in motive categories

Furthermore, open answers point towards the influence of the family situation

on the motivation to start the agritourism venture: iv) “I was close to retirement and

4.4 Results

146

physically not capable to do the hard farm labor anymore. Agritourism gave us an

opportunity to do less physical work and stay on the farm.”, v) “Previously the house

was used by the retired farming generation, after their passing the renovation was

easily affordable due to the high demand.” and vi) “Our daughter had a great interest in

running an equestrian business.”

4.4.2 Role of farmwomen in agritourism ventures

In order to examine the role of farmwomen in agritourism ventures we examined the

agritourism activities each farm is involved in, separately. Table 4.2 presents the

average involvement of farmwomen in initialization and realization of the agritourism

venture. Four interviewed agritourism farmer stated not to have a spouse, hence they

are excluded from the calculations presented in Table 4.2.

Table 4.2: Summary statistic of agritourism ventures by type and farmwomen participation

N = 13

(farms with agribusiness

and spouse)

farmwomen

initialization

farmwomen

realization

sum mean SD mean SD

Accommodation 5 2.40 0.89 3.20 1.48

Organizing events 5 2.60 1.52 2.20 0.84

Gastronomy 5 3.00 1.58 2.60 1.14

Renting out locations 4 3.50 1.00 3.00 0.00

Equestrian business 3 2.33 1.15 3.00 1.73

Agritourism 13 2.96 1.11 3.00 1.22

Note: The involvement of farmwomen in initialization and realization of the agritourism venture was

measured on a Likert scale ranging from 1 to 5 (farmwoman not involved - farmwoman mainly involved).

The minimum score of involvement was 1 and the maximum score was 5.

We find that the initialization and realization of the agritourism venture is

usually made jointly. Nevertheless, there are differences in involvement depending on

the type of agritourism venture. For farmwomen, the highest involvement with

4.4 Results

147

initialization is for renting out location ventures, with realization the highest

involvement of farmwomen is in accommodation ventures. Whereas the lowest

involvement of farmwomen is in the realization of organizing events on the farm and

the initialization of equestrian businesses.

4.4.3 Farm, farmer and household characteristics

In Table 4.3 we present a full list of variables included in our analysis. Furthermore, in

Table 4.4 we present summary statistics of farm farmer and household characteristics

for agritourism and non-agritourism farmers. Additionally we use the Mann-Whitney-

Wilcoxon Test to interpret differences amongst the two groups. We find that on

average agritourism farms are operated less frequently fulltime, furthermore they have

significantly larger portfolios of on-farm diversification activities (including a total 14

possible on-farm diversification strategies). In our sample agritourism farms are

significantly more often crop and fodder producing farms (where fodder production

represents the major income source) with an average size between 30 and 40ha. Non-

agritourism farms are more frequently horticulture and livestock farms, and on average

significantly larger (60 - 70ha). In our sample, we find significantly more female farm

operators on agritourism farms (N = 4) than on non-agritourism farms (N = 0).

Furthermore, agritourism farm operators are compared to non-agritourism farmers on

average older and hold significantly less frequently a university degree. Higher

agricultural education (e.g. certified agriculturist or masters certificate in agriculture) is

contrarily on average more likely within the group of agritourism farmers. We find

high agricultural attachment in both groups, but find a slightly stronger attachment for

4.4 Results

148

agritourism farms. In all measures of risk preferences included in our analysis we find

that agritourism farmer are more risk averse. More specifically we find that agritourism

farmer are significantly more risk averse in the multiple price list with an agricultural

decision frame and in the business statement with respect to marketing risks. On the

farm household level we find that on average agritourism farms have less family

workforce, are significantly less often married or in a long term relationship and less

often a secured successor compared to the non-agritourism farms in our sample.

4.4 Results

149

Table 4.3: Variable description

Variable description

Farm

Fulltime 1 if fulltime farming

Diversification portfolio count of diversification activities (min 0 - max. 14) a)

Horticulture main production focus on horticulture

Crop farming main production focus on crop production

Fodder producing farms main production focus on fodder production

Livestock farming main production focus on livestock production

Size For agricultural production utilized area 1 = up to 10 ha; 2

= 10 to 20 ha; 3 = 20 to 30 ha; 4 = 30 to 40 ha; 5 = 40 to

50 ha; 6 = 50 to 60 ha; 7 = 60 to 70 ha; 8 = 70 ha and

above

Farmer

Gender 1 if male

Age years

Higher general education 1 if education at university level

Higher agricultural education 1 if certified agriculturist or masters certificate in

agriculture

Attachment to agriculture scale from 1 to 5, 1 = “very attached” to 5 = “no

attachment”

Risk preferences (MPL) count of safe lottery choices, 1 = very risk loving; 5= risk

neutral; 9 = very risk averse

Risk preferences (self-assessment) inverse scale from 0 to 10, 0 = “not at all willing to take

risks” to 10 = “very willing to take risks” b)

Risk preferences (production) scale from 1 to 5, 1 = fully agree; 5 = don’t agree c)

Risk preferences (marketing) scale from 1 to 5, 1 = fully agree; 5 = don’t agree c)

Risk preferences (finances) scale from 1 to 5, 1 = fully agree; 5 = don’t agree c)

Risk preferences (agriculture generally) scale from 1 to 5, 1 = fully agree; 5 = don’t agree c)

Household

Family workforce number of family members working on the farm

Married 1 if married or in a long term relationship

Succession 1 = yes, succession secured; 0.5 = succession likely; 0 =

succession not intended in the near future; 0.5 =

succession unlikely; -1 = no succession secured

*Significant at the 10% level, ** Significant at the 5% level, *** significant at the 1% level

a) Including: direct marketing, processing of agricultural products, gastronomy, accommodation,

equestrian business, renting out locations, events, Christmas trees, social farming, renewable wind energy,

biogas production, photovoltaic systems, agricultural labor for others and woodwork.

4.5 Discussion and conclusion

150

b) We invert the original self-assessment scale of risk aversion to ease the interpretation. For all measures

of risk aversion higher values now indicate higher risk aversion. c) Agreement with the following four statements: I am willing to take more risks than my colleagues with

respect to production. 2. I am willing to take more risks than my colleagues with respect to marketing. 3. I

am willing to take more risks than my colleagues with respect to financial issues. 4. I am willing to take

more risks than my colleagues with respect to agriculture generally.

Table 4.4: Summary statistics of farm farmer and household characteristics for agritourism and

non-agritourism farms

Agritourism

farms

N = 17

Non-agritourism

farms

N = 16

Mann-

Whitney-

Wilcoxon-

Test

mean SD mean SD p-value

Farm

Fulltime 0.71 0.47 0.75 0.45 0.797

Diversification portfolio 2.88 1.41 1.69 0.60 0.007***

Horticulture 0.00 0.00 0.12 0.34 0.151

Crop farming 0.24 0.44 0.00 0.00 0.045**

Fodder producing farms 0.18 0.39 0.12 0.34 0.706

Livestock farming 0.35 0.49 0.50 0.52 0.412

Size 4.76 2.7 7.12 1.71 0.003***

Farmer

Gender 0.82 0.39 1.00 0.00 0.089*

Age 52.82 12.57 46.81 13.77 0.387

Higher general education 0.19 0.4 0.56 0.51 0.033**

Higher agricultural education 0.50 0.52 0.31 0.48 0.298

Attachment to agriculture 1.35 0.49 1.44 0.73 1.000

Risk preferences (MPL) 5.88 1.83 4.69 1.82 0.089*

Risk preferences (self-assessment) 4.88 1.96 4.00 1.32 0.184

Risk preferences (production) 3.53 1.18 3.25 1.06 0.503

Risk preferences (marketing) 4.06 0.90 2.81 1.05 0.002***

Risk preferences (finances) 3.88 1.22 3.81 1.22 0.865

Risk preferences (agriculture generally) 3.65 1.32 3.31 0.95 0.343

Household

Family workforce 1.71 0.94 1.91 0.93 0.594

Married 0.76 0.44 1.00 0.00 0.045**

Succession 0.26 0.69 0.38 0.56 0.747


This study uses interviews to examine determinants and motives underlying decisions

by farmers in the German region of Muenster to start agritourism ventures.


151

We identify pull motives to be dominant in the decision making process of

agritourism farmers in our sample. Farmers are mainly opportunity driven; they rate

the motive to use market opportunities (e.g. high demand in the area, niche markets) on

average highest. Thus, targeting extension and policies on market opportunities can be

an efficient strategy to enhance the uptake of agritourism activities. When further

categorizing the push/pull motives into economic, intrinsic, family and extrinsic

motives we find that overall intrinsic motives are rated as strongest determinants to

influence the diversification decision by agritourism farmers. Thus, the joy in working

within the agritourism sector and the contact with people are crucial for starting an

agritourism venture in our sample.

Our interviews revealed additional motives that are relevant. For instance,

synergetic effects of agritourism on other on-farm diversification strategies as well as

family issues have been identified as relevant. Many farms indicate that they use the

agritourism venture as advertising tool to attract visitors for the on-farm sale activity.

This is in line with other studies that find that agritourism ventures can have synergetic

relationships with other non-agricultural farm-diversification enterprises (Barbieri,

2009, Haugen and Vik, 2008). These results of our analysis are particularly important

for the here investigated farms that operate close to urban areas and thus can access

large customer basis.

Even though family motives are on average rated least important in our survey

(Table 4.1), farmers indicated in open parts of the interview that the motivation for

agritourism is often connected to the farm family life cycle. For some farmers


152

agritourism is an attractive alternative to decrease the physically hard work required in

conventional agriculture. This is supported by findings from Ollenburg and Buckley

(2007) who compare the motives of different types of agritourism farmers depending

on their age, farm type (long-term family vs. lifestyle migrants) and occupation (full-

time vs. part-time). They conclude that older full-time farmers scale down farm

activities and shift resources to a less labor-intensive way to earn income. Thus,

extension and policies could be specifically tailored to older farmer in order to enhance

the uptake of agritourism activities.

The importance of the farm family structure is further revealed in our analysis

when focusing on the involvement of farmwomen in the agritourism business. We find

that the agritourism venture is often a joint project of the farming couple (see also

Haugen and Vik, 2008, Phelan and Sharpley, 2011). However, the role of farmwomen

depends on the activity. Farmwomen are more involved in activities traditionally

performed by female household members (e.g. accommodation, gastronomy and

renting out locations) and less in adventure activities (e.g. the organization of events)

(Haugen and Vik, 2008). Furthermore, we find the initializing impulse to start an

agritourism venture is more frequently coming from the male spouse, whereas the daily

work is more often performed by the farmwomen.

Comparing objective characteristics of agritourism and non-agritourism

farmers, we find that agritourism farmers have a larger portfolio of diversification

activities. Thus, our results support the hypothesis that agritourism activities are more

likely to have a portfolio of on-farm diversification activities as there are synergistic


153

relationships between agritourism and other activities (see also Barbieri, 2009, Haugen

and Vik, 2008). Moreover, our results show that less labor-intensive farming practices

such as crop farming are associated with a larger diversification portfolio, compared to

farming types with a constant high labor demand (e.g. livestock production). Similar to

the conclusions presented by Vik and McElwee (2011) and Meraner et al. (2015), we

find that the agricultural size of agritourism farms is significantly smaller than non-

agritourism farms. Thus, we assume that farms in our sample have already shifted large

parts of their resources towards agritourism. We find in our sample, a larger share of

female main farm operators of agritourism farms. This is in line with the argument

proposed by Brandth and Haugen (2007) that agritourism work is traditionally female

and nonpaid work of farmwomen is capitalized in agritourism ventures. We find

agritourism farmers to be on average older, supporting the hypothesis that agritourism

is not something to start a very young age as starting capital requirements are high (see

also Haugen and Vik, 2008). Answers to open questions confirm additionally findings

by Benjamin and Kimhi (2006), agritourism ventures are increasingly attractive at a

very late stage in the farmers life when low labor intensity of some agritourism

ventures is determining the diversification decision. Non-agritourism farmer hold

significantly more often a university degree. Whereas higher agricultural education is

more common amongst agritourism farmers. Our results cannot confirm conclusions

drawn by Barbieri (2009) and Haugen and Vik (2008) that higher agricultural

education points towards a higher involvement of agritourism farmers with agriculture

as both groups of farmers show similar high attachment with agriculture.


154

Finally, we investigated the role of risk preference for agritourism activities.

The three different methods to elicit the farmers risk preferences included in our study

all point towards more risk averse preferences in the group of agritourism farmers.

More specifically, agritourism farmers are more risk averse with respect to agricultural

investments (MPL) and marketing, indicating that investment and marketing risks are

most severe for agritourism farmers. These results show that more risk averse

producers allocate more resources to activities with less income volatility (McNamara

and Weiss, 2005). Risk management support and extension may thus explicitly account

for such diversification activities. Furthermore, expected increases in production or

market risks, e.g. due to further liberalization of markets and due to climate change

(e.g. Chavas, 2011, Olesen et al., 2011) can cause a larger uptake of diversification

activities such as agritourism. On the household level, we find only significant

differences for the marital status between the two groups agritourism and non-

agritourism farms. Unlike other studies in the field we find that non-agritourism

farmers are more often married or in a long-term relationship. Nevertheless, as pointed

out previously for those farmers with a spouse the joint work at the farm is an

important factor influencing the decision to start an agritourism venture.

The novel contribution of this study lies in the combination of the analysis of

motives and objective farm and farmer characteristics, for the first time including

experimental methods to elicit farmers risk preferences, underlying the diversification

decision. This provides useful insights into the background of agritourism farmers.

However, this research should be built upon when exploring the possible relationship


155

of risk preferences and motives to start agritourism ventures on a larger scale.

Furthermore, additional research should focus on the role of all family members in

different stages of the farm diversification businesses. In this line of argument, next to

farmwomen the role of successors or other family members living on the farm should

be explored further. The results of this analysis can be useful for policy makers

wishing to exploit the motives and characteristics that trigger farmers to start

agritourism ventures outside conventional agriculture.

4.6 References

156

4.6 References

Bagi, Faqir S., and Richard J. Reeder. 2012. "Factors affecting farmer participation in

agritourism." Agricultural and Resource Economics Review 41 (2):189.

Barbieri, Carla. 2009. "A comparison of agritourism and other farm entrepreneurs:

Implications for future tourism and sociological research on agritourism."

Northeastern Recreation Research Symposium, Bolton Landing, NY, USA, March

2008.

Barrett, Christopher, Thomas A. Reardon, and Patrick M. Webb. 2001. "Nonfarm

income diversification and household livelihood strategies in rural Africa: concepts,

dynamics, and policy implications." Food Policy 26 (4):315-331.




Brandth, Berit, and Marit S. Haugen. 2007. "Gendered work in family farm tourism."

Journal of Comparative Family Studies 38 (3):379-393.

Busby, Graham, and Samantha Rendle. 2000. "The transition from tourism on farms to

farm tourism." Tourism Management 21 (6):635-642.

Chavas, Jean-Paul. 2011. "Agricultural policy in an uncertain world." European

Review of Agricultural Economics 38 (3):383-407.

Crosetto, Paolo, and Antonio Filippin. 2015. "A theoretical and experimental appraisal

of four risk elicitation methods." Experimental Economics 19 (3):1–29.

Di Domenico, Maria Laura, and Graham Miller. 2012. "Farming and tourism

enterprise: Experiential authenticity in the diversification of independent small-

scale family farming." Tourism Management 33 (2):285-294.

European Commission. 1990. Community Action to Promote Rural Tourism -

Communication from the Commission.



Evans, Nicholas J., and Brian W. Ilbery. 1989. "A conceptual framework for

investigating farm-based accommodation and tourism in Britain." Journal of Rural

Studies 5 (3):257-266.

4.6 References

157




Hardaker, Brian J., Ruud B. M. Huirne, Jock R. Anderson, and Gudbrand Lien. 2004.

Coping with risk in agriculture. 2 ed. Wallingford UK, Cambridge USA: CABI

publishing.



(3):321-336.




Hjalager, Anne-Mette. 1996. "Agricultural diversification into tourism: Evidence of a

European community development programme." Tourism Management 17 (2):103-

111.





Ilbery, Brian W., and Ian Bowler. 1998. From agricultural productivism to post-

productivism. London UK: Longman Ltd.

Ilbery, Brian W., Ian Bowler, Gordon Clark, Alastair Crockett, and Alastair Shaw.

1998. "Farmbased tourism as an alternative farm enterprise: A case study from the

northern Pennines, England." Regional Studies 32 (4):355–365.

King, Nigel, and Christine Horrocks. 2010. Interviews in qualitative research. London

UK: Sage.

Kostov, Philip, and John Lingard. 2003. "Risk management: a general framework for

rural development." Journal of Rural Studies 19 (4):463-476.

Lange, Andrej, Annette Piorr, Rosemarie Siebert, and Ingo Zasada. 2013. "Spatial

differentiation of farm diversification: How rural attractiveness and vicinity to cities

determine farm households’ response to the CAP." Land Use Policy 31:136–144.

4.6 References

158

Le Grand, Lodewijk, and Martine van Meekeren. 2008. Urban–rural relations: Dutch

experiences of the LEADER+ network and rural innovation in areas under strong

urban influences. Network for Rural Areas: Rurality near the city.

Lehner-Hilmer, Anita. 1999. Einstellungen der Landwirte zu selbständigen

Erwerbskombinationen: Auswirkungen auf den Strukturwandel in der

Landwirtschaft, Schriftenreihe agrarwissenschaftliche Forschungsergebnisse.

Hamburg GER: Kovač.




McElwee, Gerard. 2008. "A taxonomy of entrepreneurial farmers." International

Journal of Entrepreneurship and Small Business 6 (3):465-478.

McGehee, Nancy G., and Kyungmi Kim. 2004. "Motivation for agri-tourism

entrepreneurship." Journal of Travel Research 43 (2):161–170.

McGehee, Nancy G., Kyungmi Kim, and Gayle R. Jennings. 2007. "Gender and

motivation for agri-tourism entrepreneurship." Tourism Management 28 (1):280–

289.

McInerney, John P., and Martin Turner. 1991. Patterns, performance and prospects in

farm diversification. Exeter UK: University of Exeter Agricultural Economics Unit.

McNally, Sandra. 2001. "Farm diversification in England and Wales — what can we

learn from the farm business survey?" Journal of Rural Studies 17 (2):247-257.



(1):37-48.

Medhurst, Alison, and Robin Segrave. 2007. Why Do Farming Families Diversify?

Australian Government: Rural Industries Research and Development Corporation

07, 156.




Menapace, Luisa, Gregory Colson, and Roberta Raffaelli. 2015. "A comparison of



4.6 References

159



42:767-780.




Mishra, Ashok K., Hisham S. El‐Osta, and Carmen L. Sandretto. 2004. "Factors

affecting farm enterprise diversification." Agricultural Finance Review 64 (2):151-

166.

Nickerson, Norma, Rita J. Black, and Stephen F. McCool. 2001.

"Agritourism: motivations behind farm/ranch business diversification." Journal of

Travel Research 40 (1):19–26.

Nilsson, Per Åke 2002. "Staying on farms: An Ideological Background." Annals of

Tourism Research 29 (1):7-24.

Olesen, Jørgen E., Miroslav Trnka, Kurt Kersebaum, C., Arne O. Skjelvåg, Bernard

Seguin, Pirjo Peltonen-Sainio, Federica Rossi, Jerzy Kozyra, and Fabio Micale.

2011. "Impacts and adaptation of European crop production systems to climate

change." European Journal of Agronomy 34 (2):96-112.

Ollenburg, Claudia, and Ralf Buckley. 2007. "Stated economic and social motivations

of farm tourism operators." Journal of Travel Research 45 (4):444-452.

Oppermann, Martin. 1996. "Rural tourism in Southern Germany." Annals of Tourism

Research 23 (1):86–102.

Oppermann, Martin. 1997. "Rural tourism in Germany: Farm and rural tourism

operators." In The business of rural tourism: International perspectives, edited by

Stephen Page and Donald Getz, 108-119. London UK, Bonn GER, Boston USA:

International Thomson Business Press.

Pfeifer, Catherine, Roel A. Jongeneel, Marthijn P. W. Sonneveld, and Jetse J.

Stoorvogel. 2009. "Landscape properties as drivers for farm diversification: A

Dutch case study." Land Use Policy 26 (4):1106–1115.

Phelan, Chris, and Richard Sharpley. 2011. "Exploring Agritourism Entrepreneurship

in the UK." Tourism Planning & Development 8 (2):121-136.

R Core Team. 2016. "R: A language and environment for statistical computing."

http://www.R-project.org/.

http://www.r-project.org/

4.6 References

160

Reynaud, Arnaud, and Stéphane Couture. 2012. "Stability of risk preference measures:

results from a field experiment on French farmers." Theory and Decision 73

(2):203–221.

Segal, Gerry, Dan Borgia, and Jerry Schoenfeld. 2005. "The motivation to become an

entrepreneur." International Journal of Entrepreneurial Behavior & Research 11

(1):42-57.

Sharpley, Richard, and Adrian Vass. 2006. "Tourism, farming and diversification: An

attitudinal study." Tourism Management 27 (5):1040–1052.





Vik, Jostein, and Gerard McElwee. 2011. "Diversification and the entrepreneurial

motivations of farmers in Norway." Journal of Small Business Management

49:390–410.

Walford, Nigel. 2001. "Patterns of development in tourist accommodation enterprises

on farms in England and Wales." Applied Geography 21 (4):331-345.

Weaver, David B., and David A. Fennell. 1997. "The vacation farm sector in

Saskatchewan: a profile of operations." Tourism Management 18 (6):357-365.

4.7 Appendix 4.A

161

4.7 Appendix 4.A

Figure 4.A.1: Multiple price list (MPL) with agricultural frame

4.7 Appendix 4.A

162

The exact wording of the preceding question is translated from German as following:

“Assume that you are offered to make an agricultural investment. Here you will get

with different associated probabilities for investment A a return of 100.000 € or 80.000

€ and for investment B a return of 192.500 € or 5.000 €. You can choose in the

following table in each row between the two investment options (A or B).”

Furthermore, farmers where informed that real payouts are scaled down by 1.000€.

Risk preferences are indicated by the sum of safe choices made. In the last row clearly

choice B is dominating choice A, thus this question is merely a control question,

testing the subjects understanding of the task.

4.7 Appendix 4.A

163

164

Chapter 5

Diversification in peri-urban agriculture: a case study in the

Ruhr metropolitan region*

* Meraner, M., B. Pölling and R. Finger (currently under review)

165

Abstract

This article identifies determinants of the uptake and intensity of farm

diversification in the peri-urban Ruhr metropolitan region in Germany. Our

analysis uses a unique combination of risk perception and preferences with

elements of the farms’ geographical environment obtained from surveys and

geodata. A double hurdle model is used to analyze the diversification

decision and the decision on diversification intensity among farmers that

have decided to diversify. We find that high perception of market and price

risks as well as farmers’ past experiences increases the farmer’s probability

to seek income stabilizing on-farm non-agricultural diversification.

Keywords: on-farm non-agricultural diversification, decision

making, peri-urban agriculture

JEL classification: Q12

5.1 Introduction

166

5.1 Introduction

The importance of farm diversification to stabilize farm incomes and income risk has

rapidly grown over the last years (European Parliamentary Research Service, 2016,

McNamara and Weiss, 2005). The uptake of diversification activities is particularly

large in farming systems in urban and peri-urban areas because the proximity to large

agglomerations influences the demand for agricultural goods and services as well as

the opportunity costs of farming activities. More specifically, proximity to urban

centers increases incentives to develop new activities and valorizes the multifunctional

nature of agriculture due to increased demand, short supply chains, direct marketing

opportunities and community supported agriculture (Ilbery, 1991, Wilson, 2008,

Heimlich and Barnard, 1992). Contrarily the increased demand for land and labor

created by high population density generates high opportunity costs as well as

increased public control (Monaco et al., 2017). Furthermore, farm diversification in

peri-urban areas is considered essential from a societal point of view for the

maintenance of landscapes, the socio-economic viability of a region and ecological

functions, as well as environmentally sustainable forms of the provision of food and

other ecosystem services (Zasada, 2011, Heimlich and Barnard, 1992, Clark, Munroe,

and Mansfield, 2010). Thus, decisions made by farmers in urban and peri-urban areas

are of utmost importance for consumers, inhabitants of urban areas and policymakers

(McClintock, 2010). More specifically, a better understanding of drivers of farmers’

behavior enables designing more efficient policies supporting intended farm

diversification processes. In addition, revealing the processes and driving forces of

5.1 Introduction

167

farm diversification and development allows an incorporation of agriculture in models

of socio-spatial relationships in food systems in urban and peri-urban areas.

Different terms are used for commercial farming activities in and near cities or

wider agglomerations. Parallel to the global urban gardening movement with

predominantly social goals, commercial farming close to cities and regional urban food

systems are research topics of increasing interest (Ernwein, 2014). The definition of

the often synonymously used terms ‘urban farming’, ‘urban agriculture’, ‘urban agro-

food systems’, ‘metropolitan agriculture’, ‘urban fringe agriculture’, and ‘peri-urban

agriculture’ is heterogeneous in related research. We follow Opitz et al. (2016) and

define ‘peri-urban agriculture’ as ‘small- to large-scale agriculture that cultivates

agricultural land predominantly at the fringes of cities’ (p. 353) (see also Ernwein,

2014). Thus, peri-urban agriculture embraces all commercial farming activities within

cities or wider metropolitan areas.

Farm diversification is often characterized by a combination of multiple

diversification activities (Meraner et al., 2015, Haugen and Vik, 2008, Carter, 1998). In

the peri-urban context, specifically diversification activities on the farm that are not

related to core agricultural activities (i.e. on-farm non-agricultural diversification

activities), such as agritourism activities, the provision of services, equestrian

businesses, as well as on-farm processing and direct sales activities, are of particular

importance (Ilbery, 1991, Zasada, 2011).

Earlier research has identified the geographical environment, farm and farmer

characteristics as well as the individual frame of decision-making reference to

5.1 Introduction

168

influence the diversification decision (Mishra, Hisham, and Carmen, 2004, McNamara

and Weiss, 2005, Barnes et al., 2015). In particular, existing research shows that the

farmer’s decision making under uncertainty is based on the farmer’s individual frame

of reference i.e. her/his own reality that is based on past experiences, the subjective

perception of risks and risk preferences (Slovic, Fischhoff, and Lichtenstein, 1982,

Menapace, Colson, and Raffaelli, 2013, van Winsen et al., 2014).

This paper fills gaps in the literature in different dimensions. First, we expand

the dichotomous focus on diversification decisions of farms, but also investigate the

intensity of diversification in a peri-urban setting. Second, this study is the first to

include farmers’ past experiences, risk perception and risk preferences directly elicited

from the decision makers to explain diversification choices. Third, we present a unique

combination of socio-economic and geographical determinants of farm diversification.

Our empirical analysis addresses the ‘Ruhr Metropolis’, Germany’s largest polycentric

agglomeration that consists of 53 municipalities38

.

The remainder of the paper is structured as follows. The next section presents a

conceptual model of farm diversification that is the basis for our empirical analysis.

This is followed by a description of the data collection and case study area, including a

description of all variables used. The fourth section introduces the empirical approach

38

This survey has been subject of an earlier paper (Pölling et al. submitted) which the here

presented analysis extends in two ways. First, we focus on the farmer’s diversification

decision and diversification intensity in the peri-urban context as a risk management strategy.

Thus, we include possible farm adaptation strategies to the city (i.e. intensive horticulture

production) as factors influencing the diversification decision. Second, we include a wider set

of factors influencing the diversification decision, specifically focusing on the farmers

decision frame, geographical characteristics, farm and personal characteristics.

5.2 Conceptual model and theoretical background

169

and the fifth section reports the estimation results. The concluding remarks and

discussion are presented in the final section.


We consider farms as diversified if farm resources (land, labor or capital) are used for

activities other than production of conventional crops and livestock to generate

income. Thus, this also includes activities related to vertical integration (i.e. further

processing and on-farm marketing and retailing of agricultural products) (Ilbery, 1991,

McNally, 2001, Barnes et al., 2015, Weltin et al., 2017), which is also referred to as

on-farm non-agricultural diversification. More specifically, this includes agritourism

activities (gastronomy, accommodation, renting out facilities and recreational

activities), the provision of services (social services, land or forest services), equestrian

businesses, energy production (solar, wind or biogas energy production), processing

(plant products or animal products) and direct sales activities (on-farm shop, delivery

service, market stand, street stand, vending machine and party-service). The alternative

strategy to “diversification” is in the following called “no diversification”, i.e.

specializing in one activity rather than diversifying. Specialization can go hand in hand

with expansion of the farm business, but can also be a ‘business as usual’ strategy.

The extent to which resources are allocated to the non-agricultural income

activities is in many studies conceptually based on a farm household model of optimal

labor allocation. Built on an expected utility framework, rationally acting farm

households, are assumed to maximize their utility over consumption and leisure time


170

subject to time and budget constraints (Meyer, 2002, Weltin et al., 2017). However, the

expected utility framework has been criticized as it fails to describe observed behavior

(Menapace, Colson, and Raffaelli, 2013, Kahneman and Tversky, 1979, van Raaij,

1981). Hence there is a need to extend the traditional expected utility framework of

economic decision making to include intrinsic perceptions, attitude and value settings

(Slovic, Fischhoff, and Lichtenstein, 1982, van Raaij, 1981). Slovic, Fischhoff, and

Lichtenstein (1982) refer to the above as the individual’s frame of reference. In order

to understand the decision maker’s economic behavior, an understanding of her or his

frame of reference is required because the decision maker’s perceptual world forms the

basis for her or his choices. Figure 5.1 shows the theoretical framework adapted from

van Raaij (1981).

Figure 5.1: Determinants influencing farmer’s diversification decision adapted from van Raaij

(1981)

5.2.1 Diversification determinants

Several studies on on-farm diversification have shown the importance of the

geographical environment on the economic decision making in the peri-urban context.


171

Peri-urban areas are characterized by land fragmentation, high competition for land by

non-agricultural actors (Lovell, 2010, Pölling, Mergenthaler, and Lorleberg, 2016).

These factors hamper cost-efficient expansions in terms of land and livestock units,

and thus limit the possible developments in core agricultural activities (Lovell, 2010,

Zasada, 2011, Mok et al., 2014, Heimlich and Barnard, 1992). Proximity to urban

centers also goes alongside proximity to huge consumer markets promoting city-

oriented adjustments of farms in production, marketing, and service provision. Herein,

diversification is one key strategy farms situated in urbanized areas use to profit from

the nearby city and to evade urban pressures on primary production (Wilson, 2008,

Jarosz, 2008, Zasada et al., 2011, Monaco et al., 2017). Furthermore, soil quality is

associated with the farmer’s cropping decision and overall profitability. Farms located

on less productive soils have lower yield potentials and are exposed to a higher

farming risk and more volatile yields. Consequently, they are assumed to look for

additional income outside the main farming activity (Meraner et al., 2015, Lange et al.,

2013).

Observable farm and personal characteristics associated with on-farm non-

agricultural diversification include the farm manager’s occupation, farm type, size, the

farm manager’s age, education, available family workforce and succession. On-farm

diversification is often referred to as farm survival strategy presenting an opportunity

for farmers that want to stay full-time on the farm to earn extra income (Haugen and

Vik, 2008, McNally, 2001). A very common finding in the literature is that farm size

has a significant effect on the diversification decision. McNally (2001) and McNamara


172

and Weiss (2005) suggest that larger farms (as measured in terms of average net

income and number of livestock units) are more likely to be diversified, since they can

allocate and exploit available resources more efficiently. Contrarily, Mishra, Hisham,

and Carmen (2004) point out that larger farms (measured in terms of hectares of land)

profit from economies of scale, making a specialization strategy more likely, reducing

the uptake of diversification strategies. Furthermore, several researchers found that the

farm type (i.e. the production focus) influences the farmer’s diversification decision.

Aubry and Kebir (2013) and Zasada et al. (2011) argue that horticultural and

permanent crop farms (i.e. high value crop farms) are fragile when being dependent

from global markets’ long chains and are consequentially more likely to engage in on-

farm non-agricultural diversification. Furthermore, Meraner et al. (2015) and Haugen

and Vik (2008) find a larger probability of high value crop farmer to engage in a

portfolio of on-farm non-agricultural diversification activities, exploiting the synergetic

effects between on-farm sale and agritourism activities like gastronomy. Contrarily,

farms engaging in intensive livestock production are less likely to engage in on-farm

non-agricultural diversification activities due to external effects (e.g. odor nuisance)

(Zasada et al., 2011). Furthermore, the farmer’s age has been found to influence the

diversification decision. Younger farmers often seek to strengthen the farm business

viability by shifting excess labor, land and capital to on-farm diversification ventures

(Barbieri and Mahoney, 2009, McNamara and Weiss, 2005). Several empirical studies

reveal that a high education level of the main farm operator is positively associated

with higher income from non-agricultural professions, including on-farm


173

diversification activities requiring further training (McNamara and Weiss, 2005,

Benjamin and Kimhi, 2006). Succession of the agricultural holding is the main

mechanism of farm continuity for a family run farm business. According to the farm

family life cycle theory, Potter and Lobley (1996) and Dries, Pascucci, and Gardebroek

(2012) show that on-farm diversification is more likely to be perused when a successor

is present as consequentially creating a stronger need for additional income to support

the next generation’s family. Furthermore, the household size is an indicator of on-

farm labor availability, and an attribute that affects farm diversification. Meraner et al.

(2015) and Mishra, Hisham, and Carmen (2004) suggest that farms with excess family

labor capacities are more likely to create on-farm employment opportunities to use

them efficiently.

Besides observable characteristics and the geographical farm environment the

farmer’s decision making depends on the personal frame of reference. This includes

the perception of risks as well as past experiences and risk preferences. The subjective

risk perception of different risk sources is determined by the objective risk the decision

maker is exposed to and the subjective interpretation of risks. Consequentially, risk

perception is the combination of the probability of the occurrence of an uncertain event

and the consequential negative impact (Slovic, Fischhoff, and Lichtenstein, 1982).

Assefa, Meuwissen, and Oude Lansink (2016) and van Winsen et al. (2014) find

evidence that farmers who perceive risks to be very severe use long term survival

strategies such as diversification to cope with risky production, market, institutional

and labor market environments. Additionally, experienced past losses shape the

5.3 Data and case study area

174

farmer’s personal frame of reference (e.g. Tversky and Kahneman, 1973). Farmers that

experienced large losses in the past are assumed to avoid future risks and

consequentially reallocate farm resources to less risky activities (Menapace, Colson,

and Raffaelli, 2013). Earlier research on motives behind on-farm diversification found

that the main driver for on-farm non-agricultural diversification is the reduction of

risks associated with agricultural production (Barbieri and Mahoney, 2009, Hansson et

al., 2013). Within this tradition, risk aversion is generally associated with greater

willingness to adopt risk reducing strategies i.e. probability to diversify (van Winsen et

al., 2014). However, there is evidence that very risk averse farmers choose to shift the

agricultural risk to third parties (e.g. insurance) over on-farm diversification

(McNamara and Weiss, 2005, Meuwissen, Huirne, and Hardaker, 2001). Especially in

peri-urban regions, off-farm opportunities are assumed to be an attractive alternative

for risk averse farmers.


5.3.1 Case study area

The conducted analysis covers farms located in Germany’s largest polycentric

agglomeration ‘Ruhr Metropolis’. The Ruhr metropolitan area covers 53 municipalities

with in total more than five million inhabitants. The average population density is

above 1100 inhabitants / km². The core zone of the Ruhr metropolitan area is very

densely populated and dominated by build-up areas for settlements, industries, and

infrastructure, while the land use pattern becomes more heterogeneous outside of the


175

core zone and comprises larger shares of green areas. One third of Ruhr metropolitan

area is used for farming, which is high compared to other agglomerations of the global

North (Pölling, Mergenthaler, and Lorleberg, 2016). Farmland losses are more

pronounced where farmland is scarcest, so that most of the 500 ha which are on

average annually transformed into other land uses than agriculture are located in the

densely populated central. Like in monocentric cities, agricultural importance

successively increases outwards towards the peri-urban city fringe building the

transition zone to more rurally characterized areas. However, farmland is contained

even in the Ruhr Metropolis’ core zone due to the polycentric land use pattern. Green

corridors situated between the cities’ centers are often dominated by agricultural land

uses, although continuous urban encroachments reduce land resources for farming

steadily. Figure 5.2 illustrates the case study area39

.

39

Furthermore, the location of 132 surveyed farms is depicted. For 24 surveyed farmers no

match with the farms exact location was possible.


176

Figure 5.2: Map of the Ruhr metropolitan area and the location of 132 surveyed farms

5.3.2 Data and data collection

Primary data were collected from farm managers via a self-administered web survey in

spring 2016. We addressed the farm managers in the Ruhr metropolitan area via the

regional chamber of agriculture. We contacted all 2368 farm managers registered with

the chamber of agriculture which relates to 70% of the total farm population in the area

(IT.NRW, 2011). Before launching the web survey, we conducted 17 pre-tests with

farmers and agricultural students. The first invitation to participate was sent via e-mail

in the first week of March 2016; followed by a reminder two weeks later. The overall

response rate was 14%, half of all respondents fully completed the survey, leading to a

sample size of 156 farmers. The survey included following parts: (1) general

information on the farm (full- or part-time farming, farm type, size, and diversification


177

activities), (2) information on the farmer’s risk perception, past losses and risk

preferences, (3) personal demographic information on the farmer and household (age,

level of education, succession, household size). Additionally, publicly available geo

data was added to the web survey’s primary database. This enabled us to connect the

survey information with data on soil fertility, and distance to urban centers in 132

cases, i.e. farms.

Farmers risk perception is measured with two sets of questions related to four

main risk sources farmers are exposed to. Based on literature research as well as expert

interviews, we included market and price risks, production risks, institutional risks and

labor risks (Meuwissen, Huirne, and Hardaker, 2001, van Winsen et al., 2014).

Farmers were asked to indicate the importance of each source of risk as well as the

severity of the possible impact caused by the risk source using five point Likert scales

respectively (1 = “unimportant” to 5 = “very important” and 1 = “no impact” to 5 =

“severe impact”). Risk perception scores are calculated as an average over the

probability of the occurrence of an uncertain event and the consequential negative

impact (Slovic, Fischhoff, and Lichtenstein, 1982). Additionally, we asked farmers to

indicate whether they faced severe losses on the farm due to market, production or

institutional risks over the last five years. We consider two relatively simple, easy to

implement risk preference elicitation instruments and test their power in pertaining to

actual farmer’s diversification decisions by including. First, we use a general self-

assessment (SA) of risk preferences on an 11 point Likert scale (Dohmen et al., 2011).

However, risk attitudes are often assumed to differ over domains, i.e. decision makers


178

can be simultaneously risk seeking and risk averse in different domains (Dohmen et al.,

2011). Second, we thus use a set of four agriculture specific business statements (BS)

to elicit domain specific risk preferences. The business statements measure attitude

towards risk (i.e. market risk, production risk, institutional risk and agriculture risk in

general) relative to other farmers, thus we use the term relative risk attitude in the

remainder of this paper (Meuwissen, Huirne, and Hardaker, 2001, van Winsen et al.,

2014). A detailed description of all variables included can be found in Table 5.1.

In the further analysis the probability to diversify and the diversification

intensity are explanatory variables in a two-step decision process. In the first part of the

analysis the dependent variable is defined as a binomial choice between diversification

and no diversification. Within our sample, 79% of all farms engage in an on-farm non-

agricultural diversification activity, which is a higher share than in the total population

of all Ruhr farms40

. Generally, the sample’s high share of diversified farms is in line

with other studies in European peri-urban areas (see e.g. Zasada et al., 2011). For the

second part of the analysis, the intensity of diversification is measured as the number

of on-farm non-agricultural diversification activities the farm business is engaged in.

Farmers could choose from a set of 21 different diversification activities41

. Intensity of

40

Within the Ruhr metropolitan area participation in on-farm non-agricultural diversification

varies on municipality level between 30 to nearly 70% of all farms with a tendency to higher

shares towards the metropolitan’s centre (IT.NRW, 2011). The high share of diversified

farms in the sample is assumed to be linked to the intermediary point of access i.e. the

chamber of agriculture. About 70% of all farms in the region are registered in their data base,

originating from the EU CAP payment applications. 41

The included diversification activities are: agritourism activities (gastronomy,

accommodation, renting out facilities, recreational activities, other) social services, land or

forest services, other services, equestrian business, renewable energy production, biogas

production, other energy production, on-farm processing (plant products, animal products),


179

farm diversification ranges in the sample from 1 (28% of all diversified farms) to 8

(1% of all diversified farms) diversification activities.

On average, farms in our sample are located 7 km away from the closest city

with more than 100,000 inhabitants. We find that diversified farms are located on

average closer to urban hubs than non-diversified farms, a finding in line with findings

of other research in the field (see e.g. Zasada, 2011). The sample’s mean soil fertility,

as measured in ground points, of 56.1 is slightly above the metropolitan’s mean of 51.3

(Bodenschätzungsgesetz - BodSchätzG, 2007). Farms exploiting on-farm non-

agricultural diversification are larger (on average 60 ha) farmland than their non-

diversified counterpart (on average 41 ha). The sample’s share of full-time farms

(54%) and average farm size (55 ha) both suitably represent the region’s Agricultural

Census data of 52% and 48 ha respectively. Moreover, the share of high value crop

farms in our sample (13%) as well as the share of intensive livestock farms (27%) is in

line with Agricultural Census data (IT.NRW, 2011) (see Table 5.2). While 60% of the

diversified farms are full-time farms, this ratio reaches only about one third for the

non-diversified farms. Furthermore, the Agricultural Census 2010 reveals that two

thirds of the farms in the region are run by farm managers older than 45 years. This is

in line with our sample, in which the diversified farm managers are slightly younger

(49 years) than the non-diversified farm managers (54 years).

Farmers in our sample perceive institutional risks to be most severe. This

perception is on average higher within the group of non-diversified farmers compared

direct marketing (on-farm shop, delivery service, market stand, street stand, vending machine,

party-service, other).


180

to their diversified colleagues. On average, 50% of all surveyed farmer suffered from

severe losses due to market risks (e.g. volatile input and output prices) in the last five

years. Furthermore, we find that farmers in our sample are on average risk averse (for

both risk elicitation approaches), which is in line with previous findings, e.g. by

Meuwissen, Huirne, and Hardaker (2001) and Menapace, Colson, and Raffaelli (2013).

Farmers without an on-farm non-agricultural diversification strategy are on average

more risk averse than farmers engaging in diversification.


181

Table 5.1: Description of explanatory variables

Variable Description

DistUrb Distance in kilometers between farm and outward boundary of densely

built-up urban areas (i.e. city > 100,000 inhabitants)

SoilQual Soil quality (0 = very poor fertility and 100 = very good fertility)42

Size ha of cultivated land

HighVCrops 1 if high value crops producing farm

IntLivestock 1 if intensive livestock farm

Full-time 1 if full-time farmer

Age Years

Educ Highest degree of education in the German schooling system (1-9)43

Succession 1 if succession is sure, 0.5 if succession is quite sure, 0 if succession is

not planned in the next 15 years, -0.5 if succession is quite unsure, -1

if succession is unsure, -2 if farm exit is planned

HHS Number of household members

PercMarkRisk Perceived probability of occurrence (1 = “not likely”, 5 = “very

likely”) multiplied by perceived impact on 1 to 5 scale (1 = “no

impact”, 5 = “very strong impact”) of market and price risks

PercInstRisk Perceived probability of occurrence (1 = “not likely”, 5 = “very


impact”, 5 = “very strong impact”) of institutional risks

PercProdRisk Perceived probability of occurrence (1 = “not likely”, 5 = “very


impact”, 5 = “very strong impact”) of production risks

PercLabRisk Perceived probability of occurrence (1 = “not likely”, 5 = “very


impact”, 5 = “very strong impact”) of labor risks

LossMarkRisk 1 if farm business suffered from severe losses in the past five years

due to market risks

LossProdRisk 1 if farm business suffered from severe losses in the past five years

due to production risks

LossInstRisk 1 if farm business suffered from severe losses in the past five years

due to institutional risks

RA General risk preferences on scale from 0 (= willing to take risks) to 10

(= very unwilling to take risks)

Willingness to take more risks than my colleagues with respect to

RAAgric …agriculture in general, on scale from 1 = agree, 5 = don't agree.

RAMark …market and pricing, on scale from 1 = agree, 5 = don't agree.

RAProd …production, on scale from 1 = agree, 5 = don't agree.

RAFin …financial issues, on scale from 1 = agree, 5 = don't agree.

42

According to German soil evaluation law (Bodenschätzungsgesetz - BodSchätzG, 2007). 43

Education levels are: 1 = no degree, 2 = secondary school certificate (9 years), 3 = secondary

school certificate (11 years), 4 = advanced technical college certificate, 5 = high school

diploma, 6 = completed vocational training, 7 = certified manager, 8 = certified agriculturist

and 9 = masters certificate in agriculture.

5.4 Econometric model specification

182

Table 5.2: Summary statistics of explanatory variables

Variable

Full sample

N = 156

No diversification

N = 32

Diversification

N = 124

Mann-

Whitney

U-test

N mean SD N mean SD N mean SD p-value

DistUrb 132 7.19 5.79 26 9.35 6.74 106 6.66 5.44 0.0253

SoilQual 155 56.08 12.91 31 56.87 12.04 124 55.89 13.16 0.6658

Size 154 55.41 55.73 32 41.19 46.88 122 59.14 57.41 0.0304

HighVCrops 156 0.13 0.34 32 0.03 0.18 124 0.15 0.36 0.0672

IntLivestock 156 0.27 0.44 32 0.28 0.46 124 0.27 0.44 0.8662

Full-time 156 0.54 0.50 32 0.34 0.48 124 0.60 0.49 0.0107

Age 144 50.40 10.68 28 54.29 10.17 116 49.47 10.63 0.0966

Educ 153 6.17 1.95 31 6.13 2.31 122 6.18 1.85 0.6645

Succession 144 0.10 0.88 30 -0.30 1.07 114 0.21 0.79 0.0171

HHS 155 3.66 1.56 32 3.25 1.59 123 3.76 1.55 0.0749

PercMarkRisk 156 2.72 1.47 32 2.45 1.25 124 2.79 1.52 0.3484

PercInstRisk 156 2.78 1.36 32 2.91 1.31 124 2.75 1.38 0.5832

PercProdRisk 156 2.38 1.74 32 2.38 1.75 124 2.38 1.75 0.9064

PercLabRisk 156 1.19 1.43 32 0.97 1.33 124 1.25 1.46 0.1729

LossMarkRisk 156 0.55 0.50 32 0.59 0.50 124 0.54 0.50 0.5909

LossProdRisk 156 0.17 0.38 32 0.16 0.37 124 0.18 0.38 0.7810

LossInstRisk 156 0.42 0.49 32 0.41 0.50 124 0.42 0.50 0.8957

RA 156 6.10 2.35 32 7.19 1.69 124 5.81 2.41 0.0037

RAAgric 156 3.37 1.07 32 3.66 0.97 124 3.30 1.09 0.1000

RAMark 156 3.27 1.12 32 3.41 0.91 124 3.23 1.17 0.3947

RAProd 156 3.24 1.09 32 3.56 0.91 124 3.16 1.11 0.0443

RAFin 156 3.67 1.23 32 4.03 1.12 124 3.57 1.24 0.0543


We assume that each farmer has perfect discrimination capability between the risks of

different strategic choices, so that the strategy chosen by each farmer to maximize

individual utility as outlined in section 2. Reflecting the decision making process with

respect to diversification, we empirically investigate two steps: i) a discrete choice to

engage in on-farm diversification activities or not, ii) the choice of intensity (i.e. how

many different diversification activities are carried out). We assume that the

determinants at both steps do not necessarily have to be identical and equally

important. The first step is in the Poisson hurdle model referred to as an equation of


183

participation and the second is referred to as a model of event count that is conditioned

on the outcome of the first decision (Greene, 2002).

Based on these assumptions we define the underlying unobservable utility

function (𝑦𝑖∗) of the i-th farmer as a linear function of farm, farmer and geographical

characteristics (𝑋𝑖). The first step can be modeled as a binary choice of the farmer to

diversify or not to diversify. This implies that each farmer 𝑖 chooses to diversify 𝑗 = 1

or not to diversify 𝑗 = 0 depending on the observed characteristics 𝑋𝑖. Hence the

probability that the farm is diversified is given by:

𝑦𝑖,𝑗∗ = 𝛼𝑗𝑋𝑖 + 𝜀𝑖,𝑗 ∀ 𝑖 = 1, … 𝑁; 𝑗 = (0,1) (1)

Where 𝛼𝑗 is the vector of estimated coefficients associated with the diversification

decision and 𝜀𝑖,𝑗 is the unobservable error term. The first hurdle, estimating

determinants for the farmers decision to diversify or not, was modeled using a binary

logit model. The diversified farmer’s decision on how many activities to engage in can

be modeled as a truncated count model with possible outcomes ranging from minimum

1 to a maximum of K diversification activities. We assume again an underlying

unobservable utility function (𝑢𝑖∗) of the i-th farmer as a linear function of farm, farmer

and geographical characteristics (𝑍𝑖):

𝑢𝑖,𝑗∗ = 𝛽𝑗𝑍𝑖 + 𝜐𝑖,𝑗 ∀ 𝑖 = 1, ⋯ 𝑁; ∀ 𝑗 = 1, ⋯ 𝐾 (2)

Where 𝛽𝑗 is the vector to be estimated, 𝜐𝑖,𝑗 is the unobservable error term. To estimate

equations 1 and 2, we have chosen a Poisson hurdle model. It was chosen over

alternative models dealing with excess zeros (e.g. zero-inflated count model or

Heckman model), because the nature of zeros (no diversification) is unlikely due to

5.5 Results and discussion

184

sampling but an alternative farm management strategy. Furthermore, contrary to the

alternative Tobit model, the hurdle model acknowledges that the decision to diversify

and diversification intensity are determined by different processes (Ricker-Gilbert,

Jayne, and Chirwa, 2011). This is plausible for the analysis, since factors such as risk

preference affect the choice to diversify possibly different than the choice of

diversification intensity. Likelihood ratio (LR) was used in χ2 tests to compare the

Poisson model to the alternative negative binomial regression Model (Long, 1997,

Zeileis, Kleiber, and Jackman, 2008). We find that no over-dispersion of the data can

be detected, and the estimated coefficient, is not significantly different from zero,

suggesting that the Poisson is appropriate. Furthermore, the logit-Poisson hurdle model

is tested against the logit-negative binomial hurdle model. Comparing Akaike's

information criterion (AIC) and applying the Vuong test shows that the Poisson-hurdle

model is superior to the negative binomial hurdle model (Zeileis, Kleiber, and

Jackman, 2008). The empirical analysis is conducted using the statistical software R

(packages plyr, psych and lmtest are used) (R Core Team, 2016). All codes and data

are available from the authors upon request.


The results from the double hurdle analysis are presented in Table 5.3. In order to

evaluate which risk preference elicitation method better pertains the farmers’ behavior,

we apply the introduced double hurdle model to the two risk preferences elicitation

methods (i.e. measuring general risk attitude (see model 1) and domain specific risk


185

attitude (see model 2) separately. The results are consistent over both models, with

respect to the signs of the estimated coefficients. Nevertheless, the general self-

assessment of risk aversion performs better in explaining the farmer’s behavior in our

sample and further interpretations are hence based on model 1.

Our analysis shows no significant influence of the farms geographical

environment on neither the farmer’s decision on diversification nor the diversification

intensity. Within our sample, farms are highly diversified using the advantages of the

proximity to an urban agglomeration. Due to the high population density and

polycentric character of the case study region, it is difficult to identify the influence of

geographical characteristics in our sample. The peri-urban environment of the case

study area is predicted to have distinct effects on different on-farm non-agricultural

diversification activities. Zasada et al. (2011) highlight that for instance, direct sale and

equestrian services are of particular relevance in areas close to cities due to their

sensitive local consumer demand, while this is different for other diversification

activities, like agritourism, gastronomy and accommodation requiring a certain

distance to settlement areas for urban dwellers’ recreation in the countryside. Farm’s

producing high value crops such as horticulture or permanent crops are found to be

more likely to engage in a larger portfolio of different diversification activities. These

farm types are more likely to exploit the advantages of short supply chains, direct

marketing opportunities and the increased environmental awareness of consumers

regarding agricultural production in peri-urban areas (Zasada et al., 2011, Aubry and

Kebir, 2013, Kneafsey, 2010). Additionally, farms producing high value crops with a


186

higher probability to engage in on-farm sale activities are often combining them with

gastronomy and other agritourism activities (Haugen and Vik, 2008, Meraner et al.,

2015). Furthermore, our results show that being a full-time farmer is positively

correlated with the uptake probability of an on-farm non-agricultural diversification

activity. This is in line with findings by Haugen and Vik (2008) and McNally (2001),

concluding that on-farm non-agricultural diversification is a survival strategy used by

farmers wanting to stay on the farm and earn extra income. Like previous research by

McNamara and Weiss (2005) and Barbieri and Mahoney (2009), we find that with the

increasing age of the main farm operator, the uptake probability of on-farm non-

agricultural diversification activities decreases. Younger farmers are found to

strengthen the farm business viability and shift excess labor, land and capital to on-

farm diversification ventures. This is additionally supported by the positive effect of

succession on the diversification decision. Thus, our findings support Potter and

Lobley (1996) life cycle theory, arguing that a successor in place motivates the farm

family to build long term stability of the farm business and creating additional income

for the next generation’s family.


187

Table 5.3: Estimated results for the double hurdle model

Model 1 Model 2

Diversification choice Diversification intensity Diversification choice Diversification intensity DistUrb -0.060 (0.053) -0.002 (0.016) -0.060 (0.055) 0.000 (0.017) SoilQual -0.035 (0.028) 0.007 (0.007) -0.059 (0.031) 0.007 (0.007) Size -0.000 (0.007) 0.002 (0.002) 0.005 (0.009) 0.002 (0.002) HighVCrops 1.563 (1.384) 0.501** (0.205) 1.774 (1.330) 0.625** (0.225) IntLivestock -0.645 (0.841) -0.068 (0.195) -0.627 (0.849) -0.038 (0.197) Full-time 1.954** (0.977) 0.139 (0.211) 1.307 (1.018) 0.173 (0.216) Age -0.103*** (0.040) 0.010 (0.008) -0.078** (0.037) 0.012 (0.009) Educ -0.270 (0.172) -0.021 (0.057) -0.202 (0.176) -0.036 (0.061) Succession 0.993*** (0.384) 0.106 (0.129) 0.823** (0.382) 0.077 (0.139) HHS -0.135 (0.223) 0.010 (0.055) -0.168 (0.228) 0.010 (0.056) PercMarkRisk 0.511* (0.303) 0.059 (0.061) 0.617** (0.305) 0.041 (0.064) PercInstRisk -0.375 (0.304) -0.076 (0.075) -0.742* (0.407) -0.101 (0.078) PercProdRisk -0.147 (0.232) 0.037 (0.063) 0.033 (0.241) 0.053 (0.065) PercLabRisk -0.476* (0.276) 0.123* (0.074) -0.780** (0.324) 0.115 (0.080) LossMarkRisk -2.473** (1.037) -0.389* (0.211) -1.946** (0.969) -0.462* (0.218) LossProdRisk 1.965** (0.929) -0.080 (0.188) 1.453* (0.835) -0.182 (0.201) LossInstRisk 2.172* (1.166) 0.293 (0.233) 1.144* (1.090) 0.329 (0.246) RA -0.614*** (0.229) 0.002 (0.040) RAAgric -0.398 (0.465) 0.193 (0.136) RAMark 0.393 (0.515) -0.144 (0.122) RAProd 0.728 (0.628) 0.004 (0.156)

RAFin -0.541 (0.407) -0.100 (0.092)

Intercept 16.038*** (4.695) -0.430 (0.904) 12.125** (4.011) -0.180 (0.960)

Log-likelihood -170.1 on 38 Df -170 on 44 Df

Wald Chisq 52.289** 57.023*



188

Besides observable characteristics and the geographical farm environment, we

included elements of the farmer’s personal frame of reference (i.e. risk perception, past

experiences and risk aversion). We find that the perception of risk influences the

diversification decision as well as the diversification intensity. Farmers perceiving

market and price risks to be high are more likely to engage in on-farm non-agricultural

diversification. This finding corresponds with previous findings by Assefa,

Meuwissen, and Oude Lansink (2016) and van Winsen et al. (2014) who conclude that

farmers perceiving risks to be very severe use mainly long term survival strategies such

as diversification to cope. Contrarily, we find that farmers perceiving labor risks to be

high are less likely to diversify. These farmers are more likely to avoid employing

additional workforce that might be required for an expansion diversification activities.

In contrast, the available labor force is used in conventional agricultural activities.

When analyzing the effect of high labor risk perception in the diversification intensity

among farmers that have chosen to diversify, we find that high labor risk perception

increases the uptake probability of greater diversification intensity of on-farm

diversification strategies. Most on-farm diversification strategies are labor intensive

activities, often requiring additional hired labor, which is in turn increasing the risk

perception stemming from the additional workforce employed at the farm. We find that

the general decision to diversify as well as the diversification intensity is influenced by

the farmers’ past experiences. Farmer that have encountered major losses due to

market risks (e.g. volatile input and output prices) are less likely to engage in on-farm

non-agricultural diversification activities. The same effect is found for diversification


189

intensity. Thus, there is evidence that farmers with negative experiences due to

agricultural market risks choose a specialization strategy over on-farm non-agricultural

diversification. Furthermore, we find that farmers that experienced past losses due to

institutional risks (e.g. change of subsidies, changes in legal environments) are more

likely to engage in on-farm non-agricultural diversification. The current European

Union’s agricultural policy shifts away from a mono-functional production-oriented

support scheme towards supporting the multifunctional characteristics of agriculture

(Zasada, 2011). Thus, our finding support conclusions drawn by Weltin et al. (2017),

that more on-farm non-agricultural diversification would be pursued by farmers in a

scenario of total abolishment of the current European CAP subsidies. Additionally,

past losses due to agricultural production risks increase the farmer’s uptake probability

of on-farm non-agricultural diversification activities. For farmers experiencing weather

shocks, pests or other production related risks, on-farm non-agricultural diversification

is an option to stay on the farm but reallocate their resources away from the risky

agricultural production. Farmers revealing more risk averse preferences, elicited via

self-assessment, are significantly less likely to engage in on-farm non-agricultural

diversification. This is in line with findings by Meuwissen, Huirne, and Hardaker

(2001) and McNamara and Weiss (2005) who argue that very risk averse farmers

choose to shift the agricultural risk to third parties (e.g. insurance) over on-farm

diversification. In peri-urban areas off-farm opportunities are more attractive compared

to rural areas, thus on-farm diversification is associated with higher opportunity costs,

contributing to our result.

5.6 Conclusion

190

5.6 Conclusion

Explaining farm diversification is particularly relevant in urban and peri-urban settings

where the interrelation with non-farming actors is particularly large and thus

determines farmers’ decision making process and farm structures. Our results show

very high shares of diversified farms in the peri-urban Ruhr metropolitan region.

Policies should therefore be tailored to account for the special role of these farming

systems, with respect to, but also beyond diversification decisions. Our results have,

however, not been able to identify clear effects of different geographical determinants.

Future research should focus on more heterogeneous regions, e.g. by explicitly going

beyond the analysis of farms within the metropolitan region. Although market

opportunities are essential if farmers are to diversify, this paper shows that not all

farms are in a similar position to take advantage of existing opportunities. In particular

we show that the frame of reference is important when analyzing the farmer’s

diversification decision and intensity. We find that risk preferences determine

diversification decisions. More specifically, risk averse farmers are less likely to

engage in on-farm non-agricultural diversification activities but rather tend towards

specialization strategies. This will help further disentangle the complex decision

making process of on-farm risk management strategies. Our finding that risk

perception and the experience of severe losses in the past act as important driver of the

uptake of diversification decisions reveals that increasing climatic and market risks

coupled with an increasing likelihood of the occurrence of extreme events (e.g. due to

climate change and increasing liberalization of markets) might contribute to further

5.6 Conclusion

191

diversification of farms. This might create lock-in situations (e.g. Ding, Schoengold,

and Tadesse, 2009), as the uptake of diversification is based on large investments.

Subsequent research has to identify if the uptake of these diversification activities is a

first step into the leave of the agricultural sector, as this would have major implications

for the provision of food and other ecosystem services from the agricultural sector (e.g.

Mishra, Fannin, and Joo, 2014). Along these lines, the analysis of entry and exit into

specific diversification activities, as well as the interrelation with off-farm risk

allocation of resources should be conducted. Furthermore, it is reasonable to argue that

the farmer’s diversification decision and diversification intensity decision should be

analyzed using a longer run horizon. Thus, our findings encourage the use of panel

data when analyzing the farmer’s decision making process in future research.

5.7 References

192

5.7 References

Assefa, Tsion T., Miranda P. M. Meuwissen, and Alfons G. J. M. Oude Lansink. 2016.

"Price risk perceptions and management strategies in selected European food supply

chains: An exploratory approach." NJAS - Wageningen Journal of Life Sciences.

Aubry, Christine, and Leïla Kebir. 2013. "Shortening food supply chains: A means for

maintaining agriculture close to urban areas? The case of the French metropolitan

area of Paris." Food Policy 41:85-93.

Barbieri, Carla, and Edward Mahoney. 2009. "Why is diversification an attractive farm

adjustment strategy? Insights from Texas farmers and ranchers." Journal of Rural

Studies (25):58–66.

Barnes, Andrew P., Helena Hansson, Gordana Manevska-Tasevska, Shailesh S.

Shrestha, and Steven G. Thomson. 2015. "The influence of diversification on long-

term viability of the agricultural sector." Land Use Policy 49:404-412.




Bodenschätzungsgesetz - BodSchätzG. (2007) Gesetz zur Schätzung des

landwirtschaftlichen Kulturbodens.

Carter, Sara. 1998. "Portfolio entrepreneurship in the farm sector: indigenous growth in

rural areas?" Entrepreneurship & Regional Development 10 (1):17-32.

Clark, Jill K., Darla K. Munroe, and Becky Mansfield. 2010. "What counts as farming:

how classification limits regionalization of the food system." Cambridge Journal of

Regions, Economy and Society 3 (2):245-259.

Ding, Ya, Karina Schoengold, and Tsegaye Tadesse. 2009. "The Impact of Weather

Extremes on Agricultural Production Methods: Does drought increase adoption of

conservation tillage practices?" Journal of Agricultural and Resource

Economics:395-411.




(3):522–550.

5.7 References

193

Dries, Liesbeth, Stefano Pascucci, and Cornelis Gardebroek. 2012. "Diversification in

Italian farm systems: Are farmers using interlinked strategies?" New Medit 11

(4):7–15.

Ernwein, Marion. 2014. "Framing urban gardening and agriculture: On space, scale

and the public." Geoforum 56:77-86.



Greene, William H. 2002. Econometric analysis. 5 ed. Upper Saddle River USA:

Pearson Education, Inc.






(3):321-336.

Heimlich, Ralph E., and Charles H. Barnard. 1992. "Agricultural adaptation to

urbanization: Farm types in northeast metropolitan areas." Northeastern Journal of

Agricultural and Resource Economics 21 (1):50-60.



IT.NRW. 2011. "Agricultural Census Data 2010." Landesbetrieb Information und

Technik Nordrhein-Westfalen accessed 31.01.2016.

https://www.landesdatenbank.nrw.de/ldbnrw/online.

Jarosz, Lucy. 2008. "The city in the country: Growing alternative food networks in

Metropolitan areas." Journal of Rural Studies 24 (3):231–244.



Kneafsey, Moya. 2010. "The region in food—important or irrelevant?" Cambridge

Journal of Regions, Economy and Society:rsq012.

Lange, Andrej, Annette Piorr, Rosemarie Siebert, and Ingo Zasada. 2013. "Spatial

differentiation of farm diversification: How rural attractiveness and vicinity to cities

determine farm households’ response to the CAP." Land Use Policy 31:136–144.

https://www.landesdatenbank.nrw.de/ldbnrw/online

5.7 References

194

Long, Scott J. 1997. Regression models for categorical and limited dependent

variables. Vol. 7, Advanced quantitative techniques in the social sciences.

Thousand Oaks, USA: SAGE Publications Inc.

Lovell, Sarah Taylor. 2010. "Multifunctional urban agriculture for sustainable land use

planning in the United States." Sustainability 2 (8):2499-2522.

McClintock, Nathan. 2010. "Why farm the city? Theorizing urban agriculture through

a lens of metabolic rift." Cambridge Journal of Regions, Economy and Society 3

(2):191-207.

McNally, Sandra. 2001. "Farm diversification in England and Wales — what can we

learn from the farm business survey?" Journal of Rural Studies 17 (2):247-257.



(1):37-48.






42:767-780.




Meyer, Jack. 2002. "Expected Utility as a Paradigm for Decision Making in

Agriculture." In A Comprehensive Assessment of the Role of Risk in U.S.


Springer US.

Mishra, Ashok K., Hisham S. El‐Osta, and Carmen L. Sandretto. 2004. "Factors

affecting farm enterprise diversification." Agricultural Finance Review 64 (2):151-

166.

Mishra, Ashok K., Mathew J. Fannin, and Hyunjeong Joo. 2014. "Off-Farm Work,

Intensity of Government Payments, and Farm Exits: Evidence from a National

Survey in the United States." Canadian Journal of Agricultural Economics/Revue

canadienne d'agroeconomie 62 (2):283-306.

5.7 References

195

Mok, Hoi-Fei, Virginia G. Williamson, James R. Grove, Kristal Burry, Fiona S.

Barker, and Andrew J. Hamilton. 2014. "Strawberry fields forever? Urban

agriculture in developed countries: a review." Agronomy for Sustainable

Development 34 (1):21-43.

Monaco, Federica, Ingo Zasada, Dirk Wascher, Matjaž Glavan, Marina Pintar, Ulrich

Schmutz, Chiara Mazzocchi, Stefano Corsi, and Guido Sali. 2017. "Food

Production and Consumption: City Regions between Localism, Agricultural Land

Displacement, and Economic Competitiveness." Sustainability 9 (1):96.

Opitz, Ina, Regine Berges, Annette Piorr, and Thomas Krikser. 2016. "Contributing to

food security in urban areas: differences between urban agriculture and peri-urban

agriculture in the Global North." Agriculture and Human Values 33 (2):341-358.

Pölling, Bernd, Marcus Mergenthaler, and Wolf Lorleberg. 2016. "Professional urban

agriculture and its characteristic business models in Metropolis Ruhr, Germany."

Land Use Policy 58:366-379.

Pölling, Bernd, W. Sroka, and Marcus Mergenthaler. submitted. "Success of urban

farming’s city-adjustments and business models – Findings from a survey among

farmers in Ruhr Metropolis, Germany." Land Use Policy.

Potter, Clive, and Matt Lobley. 1996. "The farm family life cycle, succession paths and

environmental change in Britain's countryside." Journal of Agricultural Economics

47 (1-4):172-190.

R Core Team. 2016. "R: A language and environment for statistical computing."

http://www.R-project.org/.

Ricker-Gilbert, Jacob, Thomas S. Jayne, and Ephraim Chirwa. 2011. "Subsidies and

Crowding Out: A Double-Hurdle Model of Fertilizer Demand in Malawi."

American Journal of Agricultural Economics 93 (1):26-42.



Tversky, Amos, and Daniel Kahneman. 1973. "Availability: A heuristic for judging

frequency and probability." Cognitive Psychology 5 (2):207-232.


(1):1–24.



http://www.r-project.org/

5.7 References

196



Weltin, Meike, Ingo Zasada, Christian Franke, Annette Piorr, Meri Raggi, and Davide

Viaggi. 2017. "Analysing behavioural differences of farm households: An example

of income diversification strategies based on European farm survey data." Land

Use Policy 62:172-184.

Wilson, Geoff A. 2008. "From ‘weak’ to ‘strong’ multifunctionality: Conceptualising

farm-level multifunctional transitional pathways." Journal of Rural Studies 24

(3):367-383.

Zasada, Ingo. 2011. "Multifunctional peri-urban agriculture—A review of societal

demands and the provision of goods and services by farming." Land Use Policy 28

(4):639–648.

Zasada, Ingo, Christian Fertner, Annette Piorr, and Thomas Sick Nielsen. 2011. "Peri-

urbanisation and multifunctional adaptation of agriculture around Copenhagen."

Geografisk Tidsskrift-Danish Journal of Geography 111 (1):59–72.

Zeileis, Achim, Christian Kleiber, and Simon Jackman. 2008. "Regression models for

count data in R." Journal of Statistical Software 27 (8):1-25.

Decision making under risk in agriculture

Documents

Transcript of Decision making under risk in agriculture